# TriageGuard > Autonomous CI/CD PR Failure Triage & Pluggable `.triageguard.yml` MCP Hub. ## Core Safety & Performance Invariants - **Read-Only Default Branch Guardrail:** Never pushes directly to `main`, `master`, `production`, or `release`. Automated branches strictly match `^(?:fix|revert)/triage-[a-zA-Z0-9._-]+$`. - **Max 1 Auto-Retry per Commit SHA:** `TRANSIENT_FLAKE` auto-reruns are capped at 1 per `(repo, commit_sha)`. - **< 250ms Webhook SLA:** Constant-time HMAC SHA-256 verification and non-blocking async queue ingestion. - **< 2,500 Token Budget & < 0.25s Log Sanitizer:** C-level literal anchor scanning on 20-30MB logs. - **Episodic Failure Memory & Hallucination Verifier Gate:** Cross-Run Déjà Vu fingerprinting (`CHRONIC_FLAKE` detection) and AST/diff line verification. - **Sticky PR Comments & Opt-In Neutral Check Runs:** Single sticky PR comment updated in-place (``), plus optional non-blocking Check Runs (`conclusion: neutral`) with 1-click `[Rerun Failed CI]` and `[Open Revert PR]` buttons when `TRIAGE_PUBLISH_CHECK_RUN=true`. ## Key Endpoints & Showcase - [Landing Page](https://triageguard.dev/): Product overview & architecture - [Live Showcase Demo PRs](https://github.com/TriageGuard/TriageGuard-Demo/pulls): 5 open Pull Requests (`#6`-`#10`) demonstrating zero-config GitHub Actions (`#10`), Azure Pipelines + MISRA/Conan `CODE_DEFECT` (`#9`), CircleCI `INFRA_OOM_TIMEOUT` (`#6`), Buildkite `TRANSIENT_FLAKE` / `CHRONIC_FLAKE` (`#7`), and autonomous `revert/triage-502` Rollback PRs (`#8`) - [Executive ROI Dashboard](https://triageguard.dev/dashboard): Public aggregate ROI view & Bearer-authenticated tenant telemetry ledger - [Native MCP Server](https://triageguard.dev/mcp): JSON-RPC 2.0 Model Context Protocol endpoint (`GET /mcp`, `POST /mcp`) - [A2A Agent Card](https://triageguard.dev/.well-known/agent.json): Google Agent2Agent discovery manifest - [OIDC Discovery](https://triageguard.dev/.well-known/openid-configuration): Outbound OIDC identity federation discovery metadata - [OIDC JWKS](https://triageguard.dev/.well-known/jwks.json): Public RS256 JWKS for zero-secret `.triageguard.yml` customer MCP verification - [ROI Stats API](https://triageguard.dev/api/v1/stats): JSON telemetry and tenant quota metrics - [GitHub Webhook](https://triageguard.dev/webhooks/github): GitHub Actions `workflow_run` & `check_run` receiver - [GitLab Webhook](https://triageguard.dev/webhooks/gitlab): Optional GitLab CI Pipeline/Job Hook receiver - [Jenkins Webhook](https://triageguard.dev/webhooks/jenkins): Optional Jenkins Notification Plugin receiver ## MCP Tools Available at `POST /mcp` - `triage_ci_build_via_mcp`: Fetch CI failure diagnostics & metadata over MCP, sanitize to <2,500 tokens, and classify. - `rerun_ci_build_via_mcp`: Dispatch a guardrailed CI build rerun via MCP with max-1-retry-per-SHA enforcement. - `list_connected_mcp_providers`: List active CI & contextual MCP servers for a tenant or repository. - `diagnose_ci_failure_log`: Sanitize and classify a raw CI failure log with MCP & Repo AI Skills enrichment. - `get_recent_triage_outcomes`: Fetch recent CI/CD failure classifications and remediation outcomes. - `get_executive_roi_telemetry`: Fetch engineering hours saved, dollar ROI, and tenant quota status. - `list_repo_ai_skills`: Discover `AGENTS.md`, `CLAUDE.md`, and `.agents/skills/*/SKILL.md` for a repository.