anchor stringlengths 723 2.15k | positive stringlengths 153 8.2k |
|---|---|
Given the SQLite schema and the disputed field definitions, what read-only operation should I perform next to obtain the surrounding context needed to assess compatibility?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {... | Apply this capability when the next decision depends on missing surrounding context.. Capabilities: expand context. Accepts: evidence candidates, evidence. Returns: evidence. Evidence role: context expansion. Prerequisites: evidence candidate selected. Constraints: incomplete context required. Side effects: read. Argum... |
I've inspected four sources and found contradictions about access rules. What should I do next to compare the evidence and resolve the schema ambiguity, without modifying state?
Decision: route
Plan: {"active": true, "operation": "compare"}
Agent State: {"phase": "retrieval", "question_length_band": "long", "state_name... | This candidate supports controlled inspection of agreement or conflict between results.. Capabilities: compare evidence. Accepts: evidence. Returns: evidence comparison. Evidence role: evidence synthesis. Prerequisites: evidence inspected. Constraints: multiple evidence required. Side effects: none. Arguments: comparis... |
I've only seen snippets and need to join data to cover failure recovery requirements. What is the next step to inspect the full schema, not just the tempting near-duplicate, to ensure coverage?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "jo... | Use this candidate when the task requires task requirements and coverage.. Capabilities: update requirements. Accepts: evidence, agent state. Returns: governance state. Evidence role: governance update. Prerequisites: evidence inspected. Constraints: tracked requirements required. Side effects: local state write. Argum... |
I have fresh evidence from multiple sources but need to confirm sufficiency for failure recovery. What read-only operation should I run to verify whether the current evidence is enough?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": true, "operation": "latest_valu... | This operation handles whether available information is sufficient while preserving the current task state.. Capabilities: assess evidence. Accepts: evidence, governance state. Returns: assessment. Evidence role: governance assessment. Prerequisites: requirements updated. Constraints: canonical evidence set required. S... |
Given the remaining 4 steps, which operation should I perform next to retrieve the exact access rule for the identifier?
Decision: route
Plan: {"active": false, "operation": "lookup_exact"}
Agent State: {"phase": "synthesis", "question_length_band": "long", "state_name": "contradiction"}
Source State: {"available_modal... | Invoke the candidate to determine the selected result or next action from the available resources.. Capabilities: finalize selection. Accepts: assessment, governance state. Returns: terminal result. Evidence role: terminal decision. Prerequisites: requirements complete. Constraints: fresh sufficient assessment required... |
Since the previous attempt was ambiguous, what should be the next information-gathering step to resolve the schema?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": true, "operation": "lookup_exact"}
Agent ... | The operation produces a reliable result for the best next information-gathering step under current constraints.. Capabilities: plan retrieval. Accepts: agent state, text. Returns: plan. Evidence role: none. Prerequisites: objective available. Constraints: none. Side effects: local state write. Arguments: objective (st... |
What resources are available to check consistency across sources, given the disputed requirements?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "lookup_semantic"}
Agent State: {"phase": "governance", "question_length_band": "short", "state_na... | Apply this capability when the next decision depends on the available resources.. Capabilities: list sources. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: none. Side effects: read. Arguments: filters (object optional), scope (string required) |
What content should be examined next to satisfy the telemetry and trace requirements?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", "step": 2}, {"action_family": "tool_use"... | Use this candidate when the task requires the relevant content.. Capabilities: search content, retrieve passages. Accepts: text, pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: indexed sources required. Side effects: read. Arguments: limit (integer req... |
Given the stale assessment and known SQLite source, what exact access rule name or permission token should be extracted next?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", ... | Use this candidate when the task requires an exact name, code, or token.. Capabilities: exact pattern search. Accepts: text, code. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: textual source required. Side effects: read. Arguments: case sensitive (boolea... |
Given the stale assessment and the schema-bound query, what resource metadata or filter should be pulled to identify the relevant technical instruction?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "to... | This operation handles resource metadata and filters while preserving the current task state.. Capabilities: search metadata. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: metadata catalog required. Side effects: read. Arguments: filters (object required), s... |
To complete field definitions and compatibility checks, what fields, types, or structure should be retrieved from the Excel source?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "w... | Invoke the candidate to determine fields, types, and structure from the available resources.. Capabilities: inspect structured schema. Accepts: csv, excel, sqlite. Returns: schema. Evidence role: structure discovery. Prerequisites: source selected. Constraints: structured source required. Side effects: read. Arguments:... |
Given the disputed requirements and multi-source inspection, which matching records should be retrieved to verify consistency across sources?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "r... | The operation produces a reliable result for matching records under current constraints.. Capabilities: search structured records. Accepts: csv, excel, sqlite. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: schema known. Constraints: structured source required. Side effects: read. Argu... |
Given the stale assessment and disputed requirements, which read-only operation can inspect PDF section landmarks without side effects, preferring lexical match over semantic?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}... | This candidate supports controlled inspection of sections and structural landmarks.. Capabilities: inspect document structure, discover sections. Accepts: pdf. Returns: document structure. Evidence role: structure discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Arguments... |
We've only seen snippets from 8 mixed-format sources, but haven't done any search yet. What metadata-filtered lookup can identify the relevant pages or ranges for the acronym-heavy technical instruction?
Decision: route
Plan: {"active": false, "operation": "lookup_exact"}
Agent State: {"phase": "source_discovery", "que... | This candidate supports controlled inspection of the relevant pages or ranges.. Capabilities: search document pages. Accepts: pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Arguments: page range (array required... |
Instead of a broad semantic enumeration, which specific read operation should we use to retrieve the selected resource's telemetry and trace requirements, given we have no evidence yet?
Decision: route
Plan: {"active": true, "operation": "enumerate_records"}
Agent State: {"phase": "planning", "question_length_band": "l... | Use this candidate when the task requires the selected resource.. Capabilities: read content. Accepts: text, code. Returns: content. Evidence role: evidence inspection. Prerequisites: source selected. Constraints: local source required. Side effects: read. Arguments: offset (integer optional), resource (string required... |
To verify the failure recovery result, which read operation can compare complementary sources for symbols and definitions, given we have snippets and a near-duplicate obstacle?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1... | This operation handles symbols and definitions while preserving the current task state.. Capabilities: inspect code structure. Accepts: code. Returns: content, code structure. Evidence role: evidence inspection. Prerequisites: source selected. Constraints: code source required. Side effects: read. Arguments: depth (int... |
We have limited steps left; which candidate result should we inspect next to resolve the contradiction?
Decision: route
Plan: {"active": false, "operation": "contradiction_resolution"}
Agent State: {"phase": "governance", "question_length_band": "long", "state_name": "insufficient"}
Source State: {"available_modalities... | Invoke the candidate to determine a candidate result from the available resources.. Capabilities: inspect evidence. Accepts: evidence candidates. Returns: evidence. Evidence role: evidence inspection. Prerequisites: evidence candidate selected. Constraints: candidate evidence required. Side effects: read. Arguments: cl... |
The previous attempt was incomplete; what missing surrounding context should we retrieve now to satisfy the validation constraints?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "w... | Apply this capability when the next decision depends on missing surrounding context.. Capabilities: expand context, resolve split context. Accepts: evidence candidates, evidence. Returns: evidence. Evidence role: context expansion. Prerequisites: evidence candidate selected. Constraints: incomplete context required. Si... |
Among the results we've inspected, which one should we compare next to see if they agree or conflict on the telemetry requirements?
Decision: route
Plan: {"active": false, "operation": "lookup_semantic"}
Agent State: {"phase": "inspection", "question_length_band": "long", "state_name": "partial_evidence"}
Source State:... | Apply this capability when the next decision depends on agreement or conflict between results.. Capabilities: compare evidence. Accepts: evidence. Returns: evidence comparison. Evidence role: evidence synthesis. Prerequisites: evidence inspected. Constraints: multiple evidence required. Side effects: none. Arguments: c... |
What is the next action to cover the task requirements and check field definitions for compatibility?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "join"}
Agent State: {"phase": "governance", "question_length_band": "long", "state_name": "ins... | This candidate supports controlled inspection of task requirements and coverage.. Capabilities: update requirements. Accepts: evidence, agent state. Returns: governance state. Evidence role: governance update. Prerequisites: evidence inspected. Constraints: tracked requirements required. Side effects: local state write... |
Do we have enough evidence to verify consistency across the sources?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": true, "operation": "enumerate_records"}
Agent State: {"phase": "retrieval", "question_length_band": "short", "state_name": "retrieval"}
Source State... | Use this candidate when the task requires whether available information is sufficient.. Capabilities: assess evidence. Accepts: evidence, governance state. Returns: assessment. Evidence role: governance assessment. Prerequisites: requirements updated. Constraints: canonical evidence set required. Side effects: local st... |
What should we do next to resolve the conflicting technical instructions?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "contradiction_resolution"}
Agent State: {"phase": "synthesis", "question_length_band": "short", "state_name": "contradicti... | This operation handles the selected result or next action while preserving the current task state.. Capabilities: finalize selection. Accepts: assessment, governance state. Returns: terminal result. Evidence role: terminal decision. Prerequisites: requirements complete. Constraints: fresh sufficient assessment required... |
What is the best way to gather more information to check consistency?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", "step": 2}, {"action_family": "tool_use", "result": "wea... | Invoke the candidate to determine the best next information-gathering step from the available resources.. Capabilities: plan retrieval. Accepts: agent state, text. Returns: plan. Evidence role: none. Prerequisites: objective available. Constraints: none. Side effects: local state write. Arguments: objective (string req... |
Which available resources can help compare versions for compatibility?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": true, "operation": "compare"}
Agent State: {"phase": "retrieval", "question_length_ban... | The operation produces a reliable result for the available resources under current constraints.. Capabilities: list sources. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: none. Side effects: read. Arguments: filters (object optional), scope (string required) |
Based on the inspected multi-source evidence and stale assessment, which read-only operation on mixed sources should I perform next to retrieve the relevant content for event delivery and retry behavior?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", ... | Apply this capability when the next decision depends on the relevant content.. Capabilities: search content. Accepts: text, pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: indexed sources required. Side effects: read. Arguments: limit (integer optional... |
After reviewing the full context and noting schema ambiguity, which read-only operation should I run next to extract the exact resource identifier from the code-related sources, since the assessment is missing?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false,... | This candidate supports controlled inspection of an exact name, code, or token.. Capabilities: exact pattern search. Accepts: text, code. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: textual source required. Side effects: read. Arguments: case sensitive ... |
Given the split context and unknown source inventory, should I enumerate all available PDF sources to gather metadata and filters, rather than inspecting a single passage?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"... | Use this candidate when the task requires resource metadata and filters.. Capabilities: search metadata. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: metadata catalog required. Side effects: read. Arguments: filters (object required), sort (string optional) |
To verify the result for field definitions and compatibility, which read-only operation should I perform next to enumerate sources and obtain the schema structure, given the stale assessment and complementary evidence?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family... | This operation handles fields, types, and structure while preserving the current task state.. Capabilities: inspect structured schema. Accepts: csv, excel, sqlite. Returns: schema. Evidence role: structure discovery. Prerequisites: source selected. Constraints: structured source required. Side effects: read. Arguments:... |
We have 8 steps left and one requirement unresolved. Which operation should we prioritize next to match records from the CSV?
Decision: route
Plan: {"active": false, "operation": "compatibility"}
Agent State: {"phase": "inspection", "question_length_band": "long", "state_name": "partial_evidence"}
Source State: {"avail... | The operation produces a reliable result for matching records under current constraints.. Capabilities: search structured records, filter records. Accepts: csv, excel, sqlite. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: schema known. Constraints: structured source required. Side eff... |
Previous attempt was incomplete. We need to locate sections and structural landmarks in the PDF. What should we do next?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", "step... | The operation produces a reliable result for sections and structural landmarks under current constraints.. Capabilities: inspect document structure. Accepts: pdf. Returns: document structure. Evidence role: structure discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Argume... |
Given the mixed sources, which pages or ranges are relevant for access rules? What operation should we perform next?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": false, "operation": "latest_value_select... | Apply this capability when the next decision depends on the relevant pages or ranges.. Capabilities: search document pages. Accepts: pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Arguments: page range (array r... |
We need to resolve contradictions across sources. What should we do next to select the right resource?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": false, "operation": "contradiction_resolution"}
Agent ... | This candidate supports controlled inspection of the selected resource.. Capabilities: read content. Accepts: text, code. Returns: content. Evidence role: evidence inspection. Prerequisites: source selected. Constraints: local source required. Side effects: read. Arguments: offset (integer optional), resource (string r... |
We have complete requirements and no unresolved items, but we haven't verified the absence of the exact identifier in any source. We're in planning with one step left. What should we do next?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "w... | This operation handles symbols and definitions while preserving the current task state.. Capabilities: inspect code structure, resolve symbols. Accepts: code. Returns: content, code structure. Evidence role: evidence inspection. Prerequisites: source selected. Constraints: code source required. Side effects: read. Argu... |
We have four observed evidence snippets but need to join them with schema-bound data from mixed sources. The assessment is stale, and we have two unresolved requirements. What's the next move?
Decision: route
Plan: {"active": false, "operation": "join"}
Agent State: {"phase": "synthesis", "question_length_band": "long"... | This operation handles a candidate result while preserving the current task state.. Capabilities: inspect evidence. Accepts: evidence candidates. Returns: evidence. Evidence role: evidence inspection. Prerequisites: evidence candidate selected. Constraints: candidate evidence required. Side effects: read. Arguments: cl... |
We're at terminal readiness with two steps left, have partial requirements, and need to find missing surrounding context for a workflow step. What should we do next?
Decision: route
Plan: {"active": false, "operation": "compatibility"}
Agent State: {"phase": "terminal_readiness", "question_length_band": "long", "state_... | Invoke the candidate to determine missing surrounding context from the available resources.. Capabilities: expand context. Accepts: evidence candidates, evidence. Returns: evidence. Evidence role: context expansion. Prerequisites: evidence candidate selected. Constraints: incomplete context required. Side effects: read... |
We have contradictory evidence from multiple inspected sources, and need to enumerate all sources to resolve the dispute. We're in synthesis with two unresolved requirements. What's the next step?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": ... | The operation produces a reliable result for agreement or conflict between results under current constraints.. Capabilities: compare evidence. Accepts: evidence. Returns: evidence comparison. Evidence role: evidence synthesis. Prerequisites: evidence inspected. Constraints: multiple evidence required. Side effects: non... |
We have complete requirements but one unresolved item. Given the known inventory and snippets, what read-only next step can join the complementary evidence without writing state?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "join"}
Agent Stat... | Apply this capability when the next decision depends on task requirements and coverage.. Capabilities: update requirements. Accepts: evidence, agent state. Returns: governance state. Evidence role: governance update. Prerequisites: evidence inspected. Constraints: tracked requirements required. Side effects: local stat... |
We've checked four sources and hit a split context. The plan is disputed but no requirements are unresolved. Is what we have sufficient, or should we read more?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_fami... | This candidate supports controlled inspection of whether available information is sufficient.. Capabilities: assess evidence. Accepts: evidence, governance state. Returns: assessment. Evidence role: governance assessment. Prerequisites: requirements updated. Constraints: canonical evidence set required. Side effects: l... |
Given the fresh assessment and two steps left, should we extract the exact instruction from the CSV evidence or instead update the plan? Only a read is allowed.
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"activ... | Use this candidate when the task requires the selected result or next action.. Capabilities: finalize selection. Accepts: assessment, governance state. Returns: terminal result. Evidence role: terminal decision. Prerequisites: requirements complete. Constraints: fresh sufficient assessment required. Side effects: local... |
To verify the contradiction across sources, what evidence-gathering step should we take next, given the identifier variation and the need to compare?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_... | This operation handles the best next information-gathering step while preserving the current task state.. Capabilities: plan retrieval. Accepts: agent state, text. Returns: plan. Evidence role: none. Prerequisites: objective available. Constraints: none. Side effects: local state write. Arguments: objective (string req... |
Given the stale assessment and partial progress, with 8 steps left, which available resource should I use next to join the evidence and resolve the near-duplicate issue?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "join"}
Agent State: {"phas... | Invoke the candidate to determine the available resources from the available resources.. Capabilities: list sources. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: none. Side effects: read. Arguments: filters (object optional), scope (string required) |
The previous attempt was incomplete and the context is partial. With only 1 step left, what should I do next to get the exact field definition from the relevant content?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"ac... | The operation produces a reliable result for the relevant content under current constraints.. Capabilities: search content. Accepts: text, pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: indexed sources required. Side effects: read. Arguments: limit (i... |
I need to find the exact name or token for access rules. Which available resource should I consult to enumerate the records, given the unknown inventory?
Decision: route
Plan: {"active": true, "operation": "enumerate_records"}
Agent State: {"phase": "planning", "question_length_band": "long", "state_name": "initial"}
S... | Apply this capability when the next decision depends on an exact name, code, or token.. Capabilities: exact pattern search. Accepts: text, code. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: textual source required. Side effects: read. Arguments: case sen... |
Given the contradictory evidence and schema ambiguity, what should I do next to inspect the resource metadata and filters?
Decision: route
Plan: {"active": false, "operation": "enumerate_records"}
Agent State: {"phase": "inspection", "question_length_band": "long", "state_name": "partial_evidence"}
Source State: {"avai... | Use this candidate when the task requires resource metadata and filters.. Capabilities: search metadata, filter sources. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: metadata catalog required. Side effects: read. Arguments: filters (object optional), sort (... |
Check the CSV schema to confirm the field names and types before we proceed.
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", "step": 2}, {"action_family": "tool_use", "result... | Use this candidate when the task requires fields, types, and structure.. Capabilities: inspect structured schema. Accepts: csv, excel, sqlite. Returns: schema. Evidence role: structure discovery. Prerequisites: source selected. Constraints: structured source required. Side effects: read. Arguments: include types (boole... |
Find the records that match the access rules we're looking for.
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": false, "operation": "contradiction_resolution"}
Agent State: {"phase": "source_discovery", "q... | This operation handles matching records while preserving the current task state.. Capabilities: search structured records. Accepts: csv, excel, sqlite. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: schema known. Constraints: structured source required. Side effects: read. Arguments: l... |
I need to see the sections and landmarks in the PDF to understand its structure.
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "enumerate_sources"}
Agent State: {"phase": "terminal_readiness", "question_length_band": "short", "state_name": "fr... | Invoke the candidate to determine sections and structural landmarks from the available resources.. Capabilities: inspect document structure. Accepts: pdf. Returns: document structure. Evidence role: structure discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Arguments: lan... |
Given the current partial progress, find the exact page range in the mixed sources that matches the entity we need.
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": false, "operation": "lookup_exact"}
Agent... | The operation produces a reliable result for the relevant pages or ranges under current constraints.. Capabilities: search document pages. Accepts: pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Arguments: page... |
Given the schema-bound query and version noise, which resource should be selected next to resolve field definitions, ensuring read-only and local state write only?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_f... | This candidate supports controlled inspection of the selected resource.. Capabilities: read content, inspect full source. Accepts: text, code. Returns: content. Evidence role: evidence inspection. Prerequisites: source selected. Constraints: local source required. Side effects: read. Arguments: offset (integer required... |
We've already checked the metadata and found a near-duplicate. Which symbol definition should we look up next, given the remaining steps and the need for a local state write?
Decision: route
Plan: {"active": false, "operation": "lookup_semantic"}
Agent State: {"phase": "terminal_readiness", "question_length_band": "lon... | This candidate supports controlled inspection of symbols and definitions.. Capabilities: inspect code structure. Accepts: code. Returns: content, code structure. Evidence role: evidence inspection. Prerequisites: source selected. Constraints: code source required. Side effects: read. Arguments: depth (integer required)... |
Instead of enumerating all records, which candidate result should we inspect next for event delivery and retry behavior, given the schema ambiguity?
Decision: route
Plan: {"active": false, "operation": "enumerate_records"}
Agent State: {"phase": "source_discovery", "question_length_band": "long", "state_name": "source_... | Use this candidate when the task requires a candidate result.. Capabilities: inspect evidence. Accepts: evidence candidates. Returns: evidence. Evidence role: evidence inspection. Prerequisites: evidence candidate selected. Constraints: candidate evidence required. Side effects: read. Arguments: claim (string optional)... |
How can we verify the resource limits and lifecycle by comparing the contradictory evidence? What missing context should we retrieve next?
Decision: route
Plan: {"active": false, "operation": "compare"}
Agent State: {"phase": "governance", "question_length_band": "long", "state_name": "insufficient"}
Source State: {"av... | This operation handles missing surrounding context while preserving the current task state.. Capabilities: expand context. Accepts: evidence candidates, evidence. Returns: evidence. Evidence role: context expansion. Prerequisites: evidence candidate selected. Constraints: incomplete context required. Side effects: read... |
Given the remaining steps, should we first verify whether the two inspected passages agree or conflict?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "lookup_exact"}
Agent State: {"phase": "inspection", "question_length_band": "long", "state_n... | Invoke the candidate to determine agreement or conflict between results from the available resources.. Capabilities: compare evidence. Accepts: evidence. Returns: evidence comparison. Evidence role: evidence synthesis. Prerequisites: evidence inspected. Constraints: multiple evidence required. Side effects: none. Argum... |
The previous attempt was incomplete; what should we do next to ensure all task requirements are covered?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": true, "operation": "enumerate_sources"}
Agent State: {"phase": "retrieval", "question_length_band": "long", "sta... | Apply this capability when the next decision depends on task requirements and coverage.. Capabilities: update requirements, track coverage. Accepts: evidence, agent state. Returns: governance state. Evidence role: governance update. Prerequisites: evidence inspected. Constraints: tracked requirements required. Side eff... |
Given the partial source inventory, is there enough information to proceed, or should we search for more?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", "step": 2}, {"action... | Apply this capability when the next decision depends on whether available information is sufficient.. Capabilities: assess evidence. Accepts: evidence, governance state. Returns: assessment. Evidence role: governance assessment. Prerequisites: requirements updated. Constraints: canonical evidence set required. Side eff... |
What should be the next action given the current evidence and remaining step?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", "step": 2}, {"action_family": "tool_use", "resul... | This candidate supports controlled inspection of the selected result or next action.. Capabilities: finalize selection. Accepts: assessment, governance state. Returns: terminal result. Evidence role: terminal decision. Prerequisites: requirements complete. Constraints: fresh sufficient assessment required. Side effects... |
What should I do next to find the resource limit info in the Excel files?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", "step": 2}, {"action_family": "tool_use", "result": ... | Use this candidate when the task requires the best next information-gathering step.. Capabilities: plan retrieval. Accepts: agent state, text. Returns: plan. Evidence role: none. Prerequisites: objective available. Constraints: none. Side effects: local state write. Arguments: objective (string required), priority (str... |
How can I resolve the contradiction about field definitions using the PDFs?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": true, "operation": "contradiction_resolution"}
Agent State: {"phase": "planning", "question_length_band": "short", "state_name": "initial"}
S... | This operation handles the available resources while preserving the current task state.. Capabilities: list sources. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: none. Side effects: read. Arguments: filters (object optional), scope (string required) |
What content do I need to join the SQLite tables for schema compatibility?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "join"}
Agent State: {"phase": "terminal_readiness", "question_length_band": "short", "state_name": "fresh_sufficient"}
So... | Invoke the candidate to determine the relevant content from the available resources.. Capabilities: search content. Accepts: text, pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: indexed sources required. Side effects: read. Arguments: limit (integer o... |
Which exact access rule token should I extract from the mixed sources to compare permissions?
Decision: route
Plan: {"active": false, "operation": "compare"}
Agent State: {"phase": "inspection", "question_length_band": "short", "state_name": "partial_evidence"}
Source State: {"available_modalities": ["text", "pdf", "cs... | Apply this capability when the next decision depends on an exact name, code, or token.. Capabilities: exact pattern search, identifier search. Accepts: text, code. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: textual source required. Side effects: read. ... |
I need to validate data quality but the assessment is stale. Can you check the metadata and filters for the source without modifying any state?
Decision: route
Plan: {"active": false, "operation": "join"}
Agent State: {"phase": "terminal_readiness", "question_length_band": "long", "state_name": "fresh_sufficient"}
Sour... | Apply this capability when the next decision depends on resource metadata and filters.. Capabilities: search metadata. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: metadata catalog required. Side effects: read. Arguments: filters (object required), sort (st... |
We've already checked snippets and found 4 evidence pieces. Now we need to verify absence of a field in the Excel source. What structure and types should we examine next?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Pla... | This candidate supports controlled inspection of fields, types, and structure.. Capabilities: inspect structured schema. Accepts: csv, excel, sqlite. Returns: schema. Evidence role: structure discovery. Prerequisites: source selected. Constraints: structured source required. Side effects: read. Arguments: include types... |
We have many near-duplicate records. Instead of doing a broad semantic search again, what exact matching should we do on the CSV to compare records?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_u... | Use this candidate when the task requires matching records.. Capabilities: search structured records. Accepts: csv, excel, sqlite. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: schema known. Constraints: structured source required. Side effects: read. Arguments: limit (integer require... |
The evidence is contradictory. How can we verify the result by inspecting the PDF's sections and landmarks, given the identifier variation?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": false, "operation... | This operation handles sections and structural landmarks while preserving the current task state.. Capabilities: inspect document structure. Accepts: pdf. Returns: document structure. Evidence role: structure discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Arguments: lan... |
With only 8 steps left, prioritize finding the exact pages or ranges from the PDF that match the schema.
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", "step": 2}, {"action_... | Invoke the candidate to determine the relevant pages or ranges from the available resources.. Capabilities: search document pages. Accepts: pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Arguments: page range (... |
The last attempt was incomplete; what next step should select the final resource using metadata filters?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": false, "operation": "latest_value_selection"}
Agent ... | The operation produces a reliable result for the selected resource under current constraints.. Capabilities: read content. Accepts: text, code. Returns: content. Evidence role: evidence inspection. Prerequisites: source selected. Constraints: local source required. Side effects: read. Arguments: offset (integer optiona... |
Which resource should we consult to get the symbols and definitions we need?
Decision: route
Plan: {"active": true, "operation": "lookup_semantic"}
Agent State: {"phase": "retrieval", "question_length_band": "short", "state_name": "retrieval"}
Source State: {"available_modalities": ["code", "metadata"], "evidence_topol... | Apply this capability when the next decision depends on symbols and definitions.. Capabilities: inspect code structure. Accepts: code. Returns: content, code structure. Evidence role: evidence inspection. Prerequisites: source selected. Constraints: code source required. Side effects: read. Arguments: depth (integer re... |
What is the next action to choose a candidate result, considering resource limits?
Decision: route
Plan: {"active": false, "operation": "latest_value_selection"}
Agent State: {"phase": "inspection", "question_length_band": "short", "state_name": "partial_evidence"}
Source State: {"available_modalities": ["code", "metad... | This candidate supports controlled inspection of a candidate result.. Capabilities: inspect evidence. Accepts: evidence candidates. Returns: evidence. Evidence role: evidence inspection. Prerequisites: evidence candidate selected. Constraints: candidate evidence required. Side effects: read. Arguments: claim (string op... |
I need to combine the data from multiple sources, but I'm missing the surrounding context for the evidence snippets I have. What should I do next to get the full picture?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"a... | Use this candidate when the task requires missing surrounding context.. Capabilities: expand context. Accepts: evidence candidates, evidence. Returns: evidence. Evidence role: context expansion. Prerequisites: evidence candidate selected. Constraints: incomplete context required. Side effects: read. Arguments: result i... |
I've checked all the sources, but I still don't see any results that match. Can you figure out if the results from different sources agree or conflict with each other?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"acti... | This operation handles agreement or conflict between results while preserving the current task state.. Capabilities: compare evidence. Accepts: evidence. Returns: evidence comparison. Evidence role: evidence synthesis. Prerequisites: evidence inspected. Constraints: multiple evidence required. Side effects: none. Argum... |
I need to make sure my analysis covers all the requirements. What's the best way to check that I haven't missed any technical instructions?
Decision: route
Plan: {"active": false, "operation": "compatibility"}
Agent State: {"phase": "terminal_readiness", "question_length_band": "long", "state_name": "fresh_sufficient"}... | Invoke the candidate to determine task requirements and coverage from the available resources.. Capabilities: update requirements. Accepts: evidence, agent state. Returns: governance state. Evidence role: governance update. Prerequisites: evidence inspected. Constraints: tracked requirements required. Side effects: loc... |
Given the current state and the need to recover from a failure, can you assess whether we have enough information to proceed, and if not, what should we do next?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_fam... | The operation produces a reliable result for whether available information is sufficient under current constraints.. Capabilities: assess evidence. Accepts: evidence, governance state. Returns: assessment. Evidence role: governance assessment. Prerequisites: requirements updated. Constraints: canonical evidence set req... |
Given the contradictory evidence and remaining steps, which read-only action should I take next to resolve the identifier variation, ensuring no state writes?
Decision: route
Plan: {"active": false, "operation": "lookup_semantic"}
Agent State: {"phase": "synthesis", "question_length_band": "long", "state_name": "contra... | Apply this capability when the next decision depends on the selected result or next action.. Capabilities: finalize selection. Accepts: assessment, governance state. Returns: terminal result. Evidence role: terminal decision. Prerequisites: requirements complete. Constraints: fresh sufficient assessment required. Side ... |
I've already inspected multiple sources and found near-duplicates. What should I compare next to distinguish them, given my assessment is stale?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": true, "operation": "compare"}
Agent State: {"phase": "planning", "questi... | This candidate supports controlled inspection of the best next information-gathering step.. Capabilities: plan retrieval. Accepts: agent state, text. Returns: plan. Evidence role: none. Prerequisites: objective available. Constraints: none. Side effects: local state write. Arguments: objective (string required), priori... |
Instead of re-enumerating all records, which exact-identifier lookup on the available Excel sources should I perform to resolve the disputed requirement?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "t... | Use this candidate when the task requires the available resources.. Capabilities: list sources. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: none. Side effects: read. Arguments: filters (object optional), scope (string required) |
How can I verify consistency across the Excel sources by extracting the relevant content using metadata filters, without modifying anything?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": false, "operatio... | Invoke the candidate to determine the relevant content from the available resources.. Capabilities: search content, retrieve passages. Accepts: text, pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: indexed sources required. Side effects: read. Argument... |
With no steps left, prioritize the operation that pinpoints the exact code or token from the CSV evidence, given the partial context and schema-bound query.
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family":... | Invoke the candidate to determine an exact name, code, or token from the available resources.. Capabilities: exact pattern search. Accepts: text, code. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: textual source required. Side effects: read. Arguments: c... |
After the ambiguous attempt, recover by comparing complementary PDF sources to resolve the disputed requirement and identify the next workflow step's resource metadata and filters.
Decision: route
Plan: {"active": false, "operation": "compare"}
Agent State: {"phase": "terminal_readiness", "question_length_band": "long"... | The operation produces a reliable result for resource metadata and filters under current constraints.. Capabilities: search metadata. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: metadata catalog required. Side effects: read. Arguments: filters (object requ... |
Without naming any tool, what catalog operation would enumerate the fields, types, and structure from the mixed sources to complete the schema definitions?
Decision: route
Plan: {"active": false, "operation": "enumerate_records"}
Agent State: {"phase": "governance", "question_length_band": "long", "state_name": "insuff... | Apply this capability when the next decision depends on fields, types, and structure.. Capabilities: inspect structured schema. Accepts: csv, excel, sqlite. Returns: schema. Evidence role: structure discovery. Prerequisites: source selected. Constraints: structured source required. Side effects: read. Arguments: includ... |
What is the next action to semantically look up matching records from the complementary sources, given the stale assessment and snippets-only inspection?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": fal... | This candidate supports controlled inspection of matching records.. Capabilities: search structured records. Accepts: csv, excel, sqlite. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: schema known. Constraints: structured source required. Side effects: read. Arguments: limit (integer ... |
Which tool should I use next to locate the structural landmarks and sections in the mixed-format evidence I've inspected?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", "ste... | Use this candidate when the task requires sections and structural landmarks.. Capabilities: inspect document structure. Accepts: pdf. Returns: document structure. Evidence role: structure discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Arguments: landmark (string optiona... |
What should I run to find the relevant pages or ranges in the PDFs for failure recovery, given the version noise?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", "step": 2}, ... | This operation handles the relevant pages or ranges while preserving the current task state.. Capabilities: search document pages. Accepts: pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Arguments: page range (... |
I need to verify version compatibility of the selected resource. Which tool can check that?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "compatibility"}
Agent State: {"phase": "terminal_readiness", "question_length_band": "short", "state_nam... | Invoke the candidate to determine the selected resource from the available resources.. Capabilities: read content. Accepts: text, code. Returns: content. Evidence role: evidence inspection. Prerequisites: source selected. Constraints: local source required. Side effects: read. Arguments: offset (integer optional), reso... |
Given the inspected evidence and complete requirements, which tool can compare the symbols and definitions across the mixed sources, while only reading?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": fals... | The operation produces a reliable result for symbols and definitions under current constraints.. Capabilities: inspect code structure. Accepts: code. Returns: content, code structure. Evidence role: evidence inspection. Prerequisites: source selected. Constraints: code source required. Side effects: read. Arguments: de... |
Given the stale assessment and partial source inventory, which candidate result should I inspect next to resolve the identifier variation, ensuring no state write occurs?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"a... | This candidate supports controlled inspection of a candidate result.. Capabilities: inspect evidence, verify provenance. Accepts: evidence candidates. Returns: evidence. Evidence role: evidence inspection. Prerequisites: evidence candidate selected. Constraints: candidate evidence required. Side effects: read. Argument... |
We've only seen snippets from four complementary sources. What remains to retrieve the missing surrounding context for the broad query, given the split-context obstacle?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"ac... | This candidate supports controlled inspection of missing surrounding context.. Capabilities: expand context. Accepts: evidence candidates, evidence. Returns: evidence. Evidence role: context expansion. Prerequisites: evidence candidate selected. Constraints: incomplete context required. Side effects: read. Arguments: r... |
Instead of assuming the schema is compatible, how should I verify the agreement or conflict between the field definitions from the inspected sources?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_... | Use this candidate when the task requires agreement or conflict between results.. Capabilities: compare evidence. Accepts: evidence. Returns: evidence comparison. Evidence role: evidence synthesis. Prerequisites: evidence inspected. Constraints: multiple evidence required. Side effects: none. Arguments: comparison (str... |
With fresh assessment but only partial context, how can I confirm that the candidate evidence covers all task requirements and is the latest value, given identifier variations?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "latest_value_select... | This operation handles task requirements and coverage while preserving the current task state.. Capabilities: update requirements. Accepts: evidence, agent state. Returns: governance state. Evidence role: governance update. Prerequisites: evidence inspected. Constraints: tracked requirements required. Side effects: loc... |
Given only two steps left and one unresolved field-definition requirement, should I proceed with what I have or gather more?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "result": "weak", "... | Invoke the candidate to determine whether available information is sufficient from the available resources.. Capabilities: assess evidence. Accepts: evidence, governance state. Returns: assessment. Evidence role: governance assessment. Prerequisites: requirements updated. Constraints: canonical evidence set required. S... |
The last attempt was ambiguous; what should I do next to finalize the workflow step?
Decision: route
Plan: {"active": false, "operation": "absence_verification"}
Agent State: {"phase": "synthesis", "question_length_band": "short", "state_name": "contradiction"}
Source State: {"available_modalities": ["excel", "metadata... | Apply this capability when the next decision depends on the selected result or next action.. Capabilities: finalize selection, emit terminal result. Accepts: assessment, governance state. Returns: terminal result. Evidence role: terminal decision. Prerequisites: requirements complete. Constraints: fresh sufficient asse... |
I need to check consistency across sources; what is the best way to start gathering information?
Decision: route
Plan: {"active": true, "operation": "join"}
Agent State: {"phase": "planning", "question_length_band": "short", "state_name": "initial"}
Source State: {"available_modalities": ["code", "metadata"], "evidence... | Apply this capability when the next decision depends on the best next information-gathering step.. Capabilities: plan retrieval. Accepts: agent state, text. Returns: plan. Evidence role: none. Prerequisites: objective available. Constraints: none. Side effects: local state write. Arguments: objective (string required),... |
What resources are available to address the telemetry and trace requirements?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": false, "operation": "lookup_exact"}
Agent State: {"phase": "source_discovery", ... | This candidate supports controlled inspection of the available resources.. Capabilities: list sources. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: none. Side effects: read. Arguments: filters (object optional), scope (string required) |
Given the current partial context and disputed requirement, which content should be retrieved next to resolve the version compatibility issue?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}, {"action_family": "tool_use", "... | Use this candidate when the task requires the relevant content.. Capabilities: search content. Accepts: text, pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: indexed sources required. Side effects: read. Arguments: limit (integer optional), query (stri... |
Based on the partial context and the need to resolve a contradiction, which exact name or code should we look up now?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": false, "operation": "contradiction_reso... | This operation handles an exact name, code, or token while preserving the current task state.. Capabilities: exact pattern search. Accepts: text, code. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: query available. Constraints: textual source required. Side effects: read. Arguments: c... |
To ensure consistency across sources, which resource metadata and filters should we use to find the relevant evidence?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}, {"action_family": "tool_use", "result": "weak", "step": 1}]
Plan: {"active": true, "operation": "contradiction_reso... | Invoke the candidate to determine resource metadata and filters from the available resources.. Capabilities: search metadata. Accepts: metadata. Returns: source list. Evidence role: source discovery. Prerequisites: none. Constraints: metadata catalog required. Side effects: read. Arguments: filters (object required), s... |
Considering the need to handle event delivery and retry behavior, what fields, types, and structure should we examine from the available sources?
Decision: route
Plan: {"active": false, "operation": "compatibility"}
Agent State: {"phase": "governance", "question_length_band": "long", "state_name": "insufficient"}
Sourc... | Apply this capability when the next decision depends on fields, types, and structure.. Capabilities: inspect structured schema, discover fields. Accepts: csv, excel, sqlite. Returns: schema. Evidence role: structure discovery. Prerequisites: source selected. Constraints: structured source required. Side effects: read. ... |
Given the stale assessment and known Excel source, what next operation should enumerate matching records without writing to local state?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "enumerate_records"}
Agent State: {"phase": "inspection", "q... | Apply this capability when the next decision depends on matching records.. Capabilities: search structured records. Accepts: csv, excel, sqlite. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: schema known. Constraints: structured source required. Side effects: read. Arguments: limit (i... |
We've checked two pieces of evidence but the requirement is disputed; what remains to resolve the contradiction using structural landmarks?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "contradiction_resolution"}
Agent State: {"phase": "gover... | This candidate supports controlled inspection of sections and structural landmarks.. Capabilities: inspect document structure. Accepts: pdf. Returns: document structure. Evidence role: structure discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Arguments: landmark (string ... |
We haven't inspected any evidence yet; should we compare pages from mixed sources or just read the metadata to determine version compatibility?
Decision: route
Plan: {"active": false, "operation": "compare"}
Agent State: {"phase": "terminal_readiness", "question_length_band": "long", "state_name": "fresh_sufficient"}
S... | Use this candidate when the task requires the relevant pages or ranges.. Capabilities: search document pages. Accepts: pdf. Returns: evidence candidates. Evidence role: candidate discovery. Prerequisites: source selected. Constraints: pdf source required. Side effects: read. Arguments: page range (array required), quer... |
To verify the field definitions from the Excel source, what operation should enumerate candidate sources without side effects?
Decision: route
History: [{"action_family": "tool_use", "result": "weak", "step": 0}]
Plan: {"active": false, "operation": "enumerate_sources"}
Agent State: {"phase": "source_discovery", "quest... | This operation handles the selected resource while preserving the current task state.. Capabilities: read content. Accepts: text, code. Returns: content. Evidence role: evidence inspection. Prerequisites: source selected. Constraints: local source required. Side effects: read. Arguments: offset (integer optional), reso... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.