Siddhantkt/support-ticket-triage

Implement persistent mission records and leaderboard across sessions

feat: enforce strictly non-binary (0, 1) scoring across environment and UI

Implement Super-Extraction and Fallback Drafting for AI metadata and logs

Enable fuzzy keyword matching for AI field assignment

Enhance AI log drafting and ensure audit log persistence

Optimize for rate limits and force AI field assignment

Fix reward rounding, clamp scores (0, 1), and enable dropdown interactivity

Fix HF 410 error with new router endpoint and add smart URL routing

Harden AI action parsing, synonyms, and field assignment directives

Fix auto-assignment, mitigation logs, intel search efficiency, and UI graphs

chore: trigger rebuild to resolve HF exit code 56

fix: implement Universal Metadata Synchronization across all action types

fix: hardened AI intelligence (normalization, truncation, and clear objectives)

fix: implement sliding window context to stay under 6k token limits

fix: enforce 'json' keyword in system prompt for API compliance

fix: allow custom model values in Defense Matrix (resolves llama-3.1-8b error)

fix: sync AI engine with Space secrets (OPENAI_API_KEY support)

feat: added multi-model support and live API_BASE_URL tuning in Defense Matrix

fix: resolve UnboundLocalError for 'raw' in auto-triage loop

fix: hardened AI engine with robust JSON parsing and tactical schema

fix: resolve UnboundLocalError by reordering handlers

fix: resolve IndentationError in app.py imports

feat: Meta PyTorch Hackathon Ultimate Uplink v2.0 — Added Neural Probe, Forensic Archive, and Defense Matrix

feat: implement Tactical Action Bar in Fleet Analytics tab

fix: log start_mission as step 1 and ensure rewards strictly > 0.0

fix: task score out of range — STEP_EPS 0.001→0.01, default reward 0.0→0.01, universal clamp

fix: redirect [THINKING]/debug lines to stderr — keep stdout spec-pure ([START]/[STEP]/[END] only)

fix: Round 1 compliance — fix [STEP]/[END] format, add HF_TOKEN guard, capture errors, ensure env.close()

fix: add def main() for OpenEnv multi-mode deployment validation

feat: dynamic leaderboard, 12-article KB, API key check, full partial-credit scoring

fix: partial-credit scoring + temperature=0.5 → varied scores per mission

fix: more ticket variety + improved agent prompt → varied scores per mission

fix: score strictly in (0,1) — store done reward only; timestamp → mission counter

fix: enforce score strictly in (0,1) — clamp all rewards to [0.01, 0.99]

fix: Returned component error — add hint_input to ALL_OUT and build_ui_dict return

fix: Gradio 6 compat — move css from gr.Blocks() to gr.mount_gradio_app()

refactor: complete app.py rewrite — all components fully functional

fix: report draft not appearing in UI

fix: restored SentinelSOC branding and synced dropdowns with mission data

fix: updated branding to FreshTriage and resolved blank dropdown fields

fix: resolved critical bugs in server/app.py and models.py

fix(ui): initialize dashboard with pre-populated history to ensure live visuals on load

fix(logic): harden AI JSON schema and injection logic to resolve 400 validation errors

refactor(docs): final rebrand of documentation and architecture overview

docs: complete technical text rebranding in documentation generator

docs: finalize naming consistency in README and build scripts

fix(imports): update __init__.py to use the new SentinelSOCEnvironment name

feat(premium): upgrade to SentinelSOC 2.0 with MITRE tactics, Integrity Gauge, and Tactical Map

refactor(names): complete project-wide transition to SentinelSOC terminology

fix(logic): implement auto-normalization for AI actions to prevent Pydantic errors