Commit History

23 days ago

  1. Fix gpt-oss-safeguard 400 on an empty system prompt; stop stale example labels hard-erroring Running

    Verified signature
    4a25d3a
  2. Gate the Mila guardrail on the MH preset tag only, not policy wording

    Verified signature
    dd580cc
  3. Add the Mila self-harm output guardrail; only include it in Run all for mental-health policies

    Verified signature
    1630c27
  4. Update workshop content samples from the sheet (new MH-06, revised MH-01/04/05)

    Verified signature
    2796397
  5. Update MH-P1 and MH-P2 from the workshop sheet (rewritten mental health policies)

    Verified signature
    cd67551
  6. Warn when a preset is added twice; number repeated category titles so PolicyLM accepts them

    Verified signature
    0b76257
  7. Put workshop policies first in the preset dropdown; label the old examples

    Verified signature
    542aa13
  8. Policy tab: make rule fields editable, add a loaded-policy status line

    Verified signature
    dc25662

24 days ago

  1. Run the model that was actually selected⁠

    Verified signature
    f8c1283
  2. Drive per-model prompt fields from the visible picker, not the hidden dropdown⁠

    juliets +1: Claude Opus 4.8 (1M context)
    93dd9e3
  3. Scope the system prompt to GPT-OSS; stop showing it for models that ignore it⁠

    juliets +1: Claude Opus 4.8 (1M context)
    b49008f
  4. Rename the Policy tab to match the sidebar ("Policy")⁠

    juliets +1: Claude Opus 4.8 (1M context)
    fa240d4
  5. Reorganize tabs, preview preset rules, cut em dashes from copy⁠

    juliets +1: Claude Opus 4.8 (1M context)
    5bf1594
  6. Run-all-models: run concurrently and stream rows as each finishes⁠

    juliets +1: Claude Opus 4.8 (1M context)
    e3e77dc
  7. Policy tab: load first; Testing tab: compare all models in one click⁠

    juliets +1: Claude Opus 4.8 (1M context)
    b945bbc
  8. Policy tab: add presets to the rule list instead of replacing it⁠

    juliets +1: Claude Opus 4.8 (1M context)
    4da6259
  9. PolicyLM: deliver preamble context on the free-form path + fold 'applies to all' clauses⁠

    juliets +1: Claude Opus 4.8 (1M context)
    2d70287
  10. One route into the policy: add anything as rules⁠

    Verified signature
    bca5542
  11. Make the live policy explicit instead of inferred⁠

    Verified signature
    72d7f4a
  12. Fix the Input column layout⁠

    Verified signature
    d42cd2d
  13. Open presets in the rule editor; name the policy under test⁠

    Verified signature
    ee3d5f9
  14. Collapsible rules with an on/off switch⁠

    Verified signature
    1577408
  15. Name the samples and policies; report a category for single-score models⁠

    Verified signature
    b346c97
  16. Model picker on the Testing tab; keep the policy across a refresh⁠

    Verified signature
    5c62d46
  17. Raise the queue concurrency limit for the workshop⁠

    Verified signature
    6cc77b7
  18. Parse written policies into PolicyLM categories⁠

    Verified signature
    a7d1398
  19. Stop the rule editor re-rendering on every keystroke⁠

    Verified signature
    2d9473a
  20. model_interface: fail loudly on unexpected PolicyLM responses⁠

    44de8b1
  21. Revert helpers.py and tab_policy.py to 716934db⁠

    Verified signature
    f81f49b
  22. Build the preset dropdown from the policy registry⁠

    Verified signature
    628b810
  23. Wire workshop policies and samples into the UI⁠

    juliets +1: Claude Opus 4.8 (1M context)
    716934d
  24. Stop tracking __pycache__; add .gitignore⁠

    juliets +1: Claude Opus 4.8 (1M context)
    eb0a0a0
  25. Switch CoPE model to cope-b-s1 (System One / SGLang)⁠

    juliets +1: Claude Opus 4.8 (1M context)
    fd25b5c
  26. Pre-load workshop policies and samples; make the policy tab model-agnostic⁠

    Verified signature
    3c569ac

27 days ago

  1. Add structured policy rules with per-rule PolicyLM scoring⁠

    juliets +1: Claude Opus 4.8 (1M context)
    5c92085

  1. Remove Qwen and Gemma models from the dropdown⁠

    juliets +1: Claude Opus 4.8 (1M context)
    d658f0b
  2. Add Shieldstral, PolicyLM, and CoPE-B classifier models⁠

    juliets +1: Claude Opus 4.8 (1M context)
    846e72b

  1. Temporarily removing CoPE from UI (#9)⁠

    juliets +1: samidh
    Verified signature
    4f51341

  1. Change cope label output to int

    Verified signature
    b466d6e
  2. Match cope output to expected model output (#8)⁠

    juliets +1: samidh
    Verified signature
    eea3b9f
  3. Custom inference parameters for cope (#7)⁠

    juliets +1: samidh
    Verified signature
    a21f087
  4. Fixed cope_token fetch (#6)⁠

    juliets +1: samidh
    Verified signature
    7b6bc0f
  5. Fixed cope endpoint base url

    Verified signature
    79545d2
  6. Fixed cope_token (#5)⁠

    juliets +1: samidh
    Verified signature
    063f8a8
  7. Add cope inference endpoint (#4)⁠

    juliets +1: samidh
    Verified signature
    58f145b
  8. Add cope inference token

    Verified signature
    0054098
  9. Add CoPE to run_test (#3)⁠

    juliets +1: samidh
    Verified signature
    6ff8ee8
  10. Add CoPE prompt template (#2)⁠

    juliets +1: samidh
    Verified signature
    43bd2c3
  11. Add CoPE to UI (#1)⁠

    juliets +1: samidh
    Verified signature
    4498250

  1. native error handling

    Yacine Jernite
    c90d7c6