Let’s automate CI failure triage

Compress 20–30 MB failing logs into <2,500 grounded tokens, catch deleted packages or auth errors outside PR diffs, and trigger safe reruns without leaving GitHub.

Connects natively with your CI pipelines, issue trackers, and observability stack ↴

  • GitHub Actions
  • Jenkins
  • CircleCI
  • Buildkite
  • Azure Pipelines
  • GitLab CI
  • Jira MCP
  • Sentry MCP
  • JFrog Artifactory

Productivity

Stop digging through 30 MB CI logs. Pinpoint verified root causes in a single sticky PR comment.

Product Walkthrough
triageguard.dev
Did you know?

<2,500 tokens

extracted from 20–30 MB CI logs in <0.25s via C-level literal prefix scanning

Webhook Ack SLA

<250ms HTTP 202

Constant-time signature verify + async worker enqueue.

Check Conclusion

Always Neutral

Opt-in 1-click actions (via TRIAGE_PUBLISH_CHECK_RUN) never block green merges.

Interactive .triageguard.yml → PR Preview

Switch presets to inspect validated zero-secret MCP configuration alongside representative PR diagnosis previews (click “View Live PR” for the live bot thread).

1. Commit zero-secret .triageguard.yml

.triageguard.yml
mcp_servers:
  jenkins:
    url: https://jenkins-mcp.acme.example/mcp
    ci_source: jenkins
    auth_mode: oidc
    capabilities:
      get_failure_diagnostics: get_build_failure_cause
      rerun_build: trigger_rebuild
  jira:
    url: https://jira-mcp.acme.example/mcp
    auth_mode: oidc
    marker_triggers:
      - marker: "*"
        tool_name: jira_get_issue
        arg_name: issue_key
        extract_pattern: '([A-Z]+-\d+)'
  helix-qac:
    url: https://qac-mcp.acme.example/mcp
    auto_discover: true

Validated by RepoMCPConfigLoader: strict HTTPS hosts, OIDC-only tokens, and allow_scm_writes=false.

2. Single Sticky PR Diagnosis + Check Run

triageguard[bot] commentedCheck Run: neutral (opt-in)
CODE_DEFECT • 95%Azure-Pipelines / ecu-static-analysis • <2,500 tokens

Root Cause: MISRA C:2012 Rule 11.3 unaligned pointer cast in src/can_bus.c (Line 42) plus missing Conan binary boost/1.86.0@corp/stable.

src/can_bus.c:42 • Additional Issues Noted: Missing Conan boost/1.86.0@corp/stable
MCP Diagnostics & Repo AI Skills Consulted (3 OIDC calls)
  • azure_devops.get_pipeline_failure_cause — fetched static analysis diagnostics via OIDC
  • jira.jira_get_issue — linked ticket blocker from PR branch metadata
  • helix-qac + AGENTS.md — loaded misra-c-compliance & prescribed aligned memcpy()
Previous TriageGuard Diagnoses on this PR (Sticky Update)
Updated single sticky comment in-place without PR notification spam.
Rerun Failed CI (opt-in)Open Revert PR (opt-in)
View Live PR #9 →

Non-diff blockers surface missing Conan, pip/uv, npm, or Docker artifacts and 401/403 auth errors alongside code bugs.

Inspect blocker coverage
Additional Issues Noted • boost/1.86.0@corp/stable missingCaught
Registry 404 / 429 rate-limit / No space left on deviceNo false blame

Neutral Check Runs expose 1-click Rerun and Open Revert PR actions inside GitHub without ever blocking green PR merges.

Explore live Demo PRs
TriageGuard Check Run — conclusion: neutralnon-blocking
branch: revert/triage-* onlymain read-only

Guardrails & Live PRs

Automate with hard safety boundaries. Inspect live PR triage in the public TriageGuard-Demo repository.

PR #9 & #10 • CODE_DEFECT

GitHub Actions + MISRA & Conan

Zero-config Actions compiler defect (#10) plus src/can_bus.c:42 & Conan blocker (#9).

PR #6 • INFRA_OOM_TIMEOUT

LTO OOM Exit 137

Separates container memory exhaustion from application logic bugs cleanly.

View PR #6 →
PR #7 • TRANSIENT_FLAKE

1-Retry Cap & Chronic Flake

Allows 1st rerun, suppresses loops, and flags CHRONIC_FLAKE.

View PR #7 →
PR #8 • CREATE_REVERT_PR

Isolated revert/triage-*

Protects main by opening automated revert PRs for human review.

View PR #8 →

More Details & Guardrails

Inspect exact architectural guarantees enforced on every webhook delivery.

In-PR Sticky Comments & Non-Blocking Check Runs
  • Updates one sticky comment in-place (<!-- triageguard-sticky-comment -->) without PR comment spam.
  • Optional neutral Check Runs (TRIAGE_PUBLISH_CHECK_RUN=true) add [Rerun Failed CI] and [Open Revert PR] buttons without blocking green merges.
  • Verifier gate checks cited files/lines against logs and added (+) diff hunks.
Detects Missing Packages, Auth, DNS & Disk Errors Outside Diffs
  • Classifies missing Conan, pip/uv, npm/pnpm, Maven/Gradle, APT & Docker 404 artifacts as CONFIG_DRIFT without blaming unrelated PR files.
  • Surfaces secondary issues (missing packages, 401/403, network/429 timeouts, No space left on device) under Additional Issues Noted.
Zero-Secret .triageguard.yml MCP & Repo AI Skills
  • GitHub Actions works out of the box; connect Jenkins, CircleCI, Buildkite, Azure, GitLab, Jira, or Sentry in .triageguard.yml using 60s signed OIDC JWTs (/.well-known/jwks.json).
  • Auto-loads AGENTS.md, CLAUDE.md, and matching .agents/skills/*/SKILL.md files per repo.
Metadata-Only Storage & Read-Only Branch Protection
  • Direct pushes to main, master, production, or release are blocked; automated branches must match ^(?:fix|revert)/triage-*.
  • Raw CI logs, diffs, and LLM prompts stay in memory only—durable storage holds operational metadata only.

Security & Native MCP

Embed zero-trust verification into triage. Short-lived 60s OIDC JWTs, metadata-only persistence, and native IDE integration.

triageguard-pipeline.yml
on: workflow_run, external_ci_webhook
HMAC + Async Enqueue<250ms
Steps
Log Sanitizer (<2,500 tok)<0.25s
Per-Request OIDC MCP Enrichment60s JWT
Verifier Gate & Sticky PR ActionNeutral
Open Standards

Connect IDE & AI Agents

mcpServers.json — Claude Code, Cursor, Antigravity
{
  "mcpServers": {
    "triageguard": {
      "type": "http",
      "url": "https://triageguard.dev/mcp"
    }
  }
}
  • Native MCP Server: /mcp

    JSON-RPC 2.0 log slicer, episodic flake memory, and safe branch validator tools.

  • Google A2A Agent Card: /.well-known/agent.json

    Machine-readable multi-agent capability discovery and skill descriptors.

  • AI Discovery Manifest: /llms.txt

    Concise architectural summary and endpoints for automated coding assistants.

Pricing

Start free on public or private repositories. Scale up cleanly with metadata-only durability and dedicated SLAs.

Free Forever

Starter

The essentials for individuals and small teams

$0

per month forever

Install Free Plan
  • 50 triages / month
  • GitHub Actions & sticky PR comments
  • Max-1-Retry & Episodic Memory
  • Repo AGENTS.md / Skills
Recommended

Pro Team

Multi-CI orchestration and live ROI observability

$99

per month per org

Request Pro Team ($99/mo)
  • 500 triages / month
  • Multi-CI .triageguard.yml OIDC MCPs
  • Native /mcp & Google A2A card
  • Slack, Teams & Executive ROI Dashboard
Dedicated SLA

Enterprise & Embedded

Private VPC deployment and custom MCP adapters

$499

per month flat

Contact Sales
  • 10,000+ triages & custom quotas
  • Custom tenant MCP adapters (QAC, Conan)
  • Self-hosted Azure / Private VPC
  • Dedicated engineering SLA

Catch failures faster on every branch

Install TriageGuard on your repositories in seconds or inspect real failing PR diagnostics running live right now.