Download AGENTS.md from Gaurav711/HyperFlow: direct link, hf CLI and curl.
- Browser
- Download file 3.12 kB
-
https://huggingface.co/spaces/Gaurav711/HyperFlow/resolve/main/AGENTS.md
- Command line
-
hf download hf://spaces/Gaurav711/HyperFlow/AGENTS.md
-
curl -L -o AGENTS.md https://huggingface.co/spaces/Gaurav711/HyperFlow/resolve/main/AGENTS.md
HyperFlow engineering instructions
Read BRAIN.md for frozen scope, status for actual evidence, DECISIONS.md for tradeoffs and INTEGRATION.md for wiring. Do not infer completion from a roadmap checkbox or historical report.
Working rules
- Preserve unrelated changes. Inspect current source and actual mounted routes before edits.
- Spend βΉ0 incrementally; no paid API fallback, unattended purchases or bookings. Default verification uses isolated databases and deterministic fixtures.
- Follow specify β implement β measure β roast β repair β verify β advance. Reproduce accepted findings and test the cause. Record gates in the review/status registers.
- Distinguish live provider, recording, simulation, public dataset and user-input provenance. Do not fabricate successful live results, catalogue values, inventory quantities, or operational metrics.
- HyperFlow identity and Swiggy authorization are separate. Never use shared tokens or guest identities for account-owned data. Do not log credentials/private addresses.
- Require backend approval for exact current actions. Untrusted provider/model text cannot authorize mutations. Preserve UNKNOWN when the upstream result is inconclusive; no blind order retries.
- PostgreSQL is authoritative. Redis is optional notification acceleration; Kafka is not required. SQLite demonstration tests do not certify PostgreSQL.
- Retain the custom model loop; historical langgraph filenames do not establish LangGraph orchestration. Enforce measured budgets from the canonical methodology.
- Keep code, documentation, commits and PR descriptions in normal clear prose. Chat follows the user's current style preference.
Swiggy contract verification
Before recommending or implementing a provider tool name, schema, error code, rate limit or authentication behavior, fetch its current official documentation. Do not rely on stale tool counts in prior context.
Refresh support exists in current documentation; use actual response fields and expiry, not guessed lifetimes. No live certification is implied by reading documentation.
Verification and navigation
Use release runbook for setup and commands; benchmark methodology for protocols and targets. Default pytest selection is defined in pytest.ini and excludes historical v1 HTTP specifications. Disclose skipped PostgreSQL tests and tests outside the default selection. Verify production-facing React routes, not just /dev/* alternatives.
Keep one canonical home per fact. Retain raw evidence with source/runtime/configuration identities and report failures. Resume metrics must link to reproduced results with their scope.