Imported from jimmypaolini/codebase (.agents/skills/triage-deployment/SKILL.md). Install upstream with npx skills add jimmypaolini/codebase --skill triage-deployment. Copyright stays with the author.

Triage CI Failures

Diagnose failing GitHub Actions workflows in this codebase, map errors to their root causes, read the relevant configuration, apply targeted fixes, and verify locally.

When to Use

• A CI check is red on a pull request or push
• The user pastes GitHub Actions log output and asks for a fix
• A gh run or workflow URL is provided for inspection
• Asked to "fix CI", "debug the failing check", or "triage CI errors"

Step 1: Obtain the Logs

Option A — Specific logs provided in context

If the user has pasted log output or given a specific run URL in $ARGUMENTS, fetch only that run and skip to Step 2:

# e.g. https://github.com/JimmyPaolini/codebase/actions/runs/12345678
gh run view 12345678 --log-failed

Option B — No logs provided: fetch ALL failing runs for the current PR

When no logs are provided, list all failing checks on the current PR and retrieve logs for each. Work through each failure in sequence — do not stop after the first fix.

# List all failing checks on the current PR (name + run URL)
gh pr checks --json name,state,link \
  --jq '.[] | select(.state == "FAILURE") | "\(.name) \(.link)"'

Parse the run ID from each link (last path segment of the URL) and fetch its logs:

gh run view <run-id> --log-failed

For each failing run, fetch its logs:

gh run view <run-id> --log-failed

Process all failing runs before moving to Step 4. Each failure may require a separate fix — apply them all before verifying.

Step 2: Identify the Workflow and Failing Job

Match the log header against the known workflows:

Identify which step within the job failed (visible in the log as ##[error] or step exit code != 0).

Step 3: Triage by Workflow

🧑‍💻 Lint Codebase — pnpm exec nx affected --target=lint-codebase

The lint-codebase target is an nx:noop whose dependsOn leaves do the work. Identify which sub-target failed:

There is no lint or format leaf, and no lint or format target anywhere in the workspace: nx affected --target=lint exits 0 printing "No tasks were run". Formatting and linting are split across oxfmt, eslint, and oxlint, plus ruff-format/ruff-lint for Python and sqlfluff-format/sqlfluff-lint for SQL. Run pnpm exec nx show project codebase --json for the full dependsOn list, which also covers callidescope, check-catalog-manifests, check-lockfile, conformetry-validate, dependency-cruiser, nbstripout, sherif, squawk, stylelint, sync-vscode-extensions, syncpack, and vulture.

Common fixes:

• typecheck: Add proper types, null checks, fix imports. Never use any — use unknown or proper typing.

  Lesson: Due to strict rules, array/object indexing (items[0]) returns undefined (use ?. or ??), any is strictly forbidden, functions must have explicit return types, and type-only imports must use import { type X } (verbatimModuleSyntax).

• eslint / oxlint: Apply the rule fix. Only use // eslint-disable-next-line when no code fix is possible.

  Lesson: ESLint v9 flat config files arrays do NOT support brace expansion. files: ['**/*.{json}'] will NOT match package.json — this silently prevents all rules in that config block from applying (e.g., @nx/dependency-checks ignoredDependencies). Use files: ['**/*.json'] (no braces) for single-extension patterns.

• oxfmt: Run pnpm exec nx affected -t oxfmt --configuration=write to auto-fix. Do NOT hand-edit formatted output.
• spell-check: Fix typos, or add legitimate technical words to configuration/cspell.config.yaml under words.

  Lesson: cspell does NOT auto-discover configuration/cspell.config.yaml (it is not in cspell's standard discovery path). The --config configuration/cspell.config.yaml flag must always be passed explicitly with workspaceRoot as CWD. Project-specific targets that run from a different CWD must use a relative path: --config ../../configuration/cspell.config.yaml.

• knip: Remove the unused export/import, or add to ignoreDependencies / ignore in configuration/knip.config.ts.
• fallow-dead-code: The same finding as knip, found workspace-wide in one pass instead of once per workspace. Fix the code and both pass. If an exception is warranted, add it to configuration/fallow.config.jsonc and to its counterpart in configuration/knip.config.ts together — the two are kept in step deliberately, so excusing it in one leaves the other still failing. Suppress a file through overrides rather than ignorePatterns: ignorePatterns drops the file from the module graph entirely, so everything it imports then reads as unreachable too.

  Lesson: String-referenced dependencies in config files (like @commitlint/config-conventional or stylelint-config-standard) are invisible to Knip and will be flagged as unused. Add them explicitly to ignoreDependencies.

• markdown-lint: Fix against configuration/.markdownlint-cli2.jsonc rules — check MD049 style (underscore), MD013 line length, and fenced code block languages.

  Lesson: If MD049 violations appear after running the formatter — oxfmt/prettier converts *emphasis* → _emphasis_. The .markdownlint-cli2.jsonc MD049 rule must use style: underscore (not asterisk) to match formatter output; using asterisk will conflict on every formatted file.

• yaml-lint: Fix indentation, trailing spaces, or document-start issues as reported.
• A synchronization target (conformetry-generators, conventional-config, devcontainer-configuration, pull-request-template, skill-exclusions): Re-run its own :write configuration and commit what it generates — never hand-edit a synchronized file.

Verify:

pnpm exec nx affected -t lint-codebase

The three sections below read as convention checks, but none of them is part of Validate Conventions — a failure in any of them shows up in a Lint Codebase run. They do not all come from the same leaf, and that decides how each one is fixed:

• Five derivation targets on the synchronization project join the same nx affected invocation as lint-codebase: conformetry-generators, conventional-config, devcontainer-configuration, pull-request-template, and skill-exclusions — each its own Nx target rather than a configuration of a shared aggregate. Only the two that fail most often are written up below; every one of them is fixed the same way, by running its own :write configuration and committing what it generates. Read the failure output to see which target reported the drift.

🏛️ Validate Convention Configuration

Failing command: npx nx run synchronization:conventional-config:check

Config: tools/synchronization/src/modules/conventional-config/conventional-config.service.ts

Fix: Run npx nx run synchronization:conventional-config:write and commit the generated changes.

📋 Validate Pull Request Template

Failing command: npx nx run synchronization:pull-request-template:check

Fix: Run npx nx run synchronization:pull-request-template:write and commit.

🧩 Validate Skill Exclusions

Failing command: npx nx run synchronization:skill-exclusions:check

Config: tools/synchronization/src/modules/skill-exclusions/skill-exclusions.constants.ts, deriving the blocks in configuration/.prettierignore, configuration/.codometerignore, .gitattributes, configuration/cspell.config.yaml, and configuration/.markdownlint-cli2.jsonc from skills-lock.json

This synchronization generates the marker-delimited block in each of the five files whose tool reaches .agents/: prettier (scans .), codometer (scans --directory .), GitHub Linguist (reads every committed file), cspell, and markdownlint. It fails when a block does not match the lockfile, which is what skills update adding a new skill would otherwise do silently.

Fix: Run npx nx run synchronization:skill-exclusions:write and commit the regenerated blocks — never hand-edit one.

🧑‍🔬 Test Coverage — nx affected --target=vitest --parallel=3 --configuration=coverage

Coverage reports are uploaded as artifacts (coverage-reports).

Triage steps:

1. Identify the failing test name and assertion error in the log output.
2. Locate the test file (*.unit.test.ts, *.integration.test.ts, *.end-to-end.test.ts).
3. Read the test and implementation side-by-side.
4. Fix the implementation logic, test expectation, or update snapshots.

Config pointers:

• Vitest base config: configuration/vitest.config.ts
• Per-project vitest config: <project>/vitest.config.ts
• Coverage thresholds: check vitest.config.ts in the failing project for coverage.thresholds

Verify:

pnpm exec nx affected -t vitest --configuration=coverage --parallel=3

🧑‍⚖️ Validate Conventions

Each step is independent. Identify which step failed:

🎋 Validate Branch Name

Config: validate-branch-name.config.cjs

Required format: <type>/<scope>-<description> (e.g., feat/lexico-user-auth)

Fix: Load and follow the checkout-branch skill to rename the branch correctly.

📝 Validate Pull Request Title

Config: configuration/commitlint.config.ts

Required format: <type>(<scope>): <gitmoji> <subject> (max 128 chars, lowercase imperative subject)

Fix: Update the PR title in the GitHub UI.

🪢 Validate Pull Request Body

Command: validation pull-request-body, the pull-request-body check

Template: .github/PULL_REQUEST_TEMPLATE.md

Required sections (exact heading text): ## 🌰 Summary, ## 📝 Details, ## 🧪 Testing, ## 🔗 Related

Two failure modes, and a body that hits both is reported against both:

• A required heading is absent — ❌ Missing required sections: 🧪 Testing — add the heading and its content.
• A template comment survives unfilled — ❌ Unfilled template comments remain: followed by one line per offending comment, such as - <!-- List of specific changes made --> — every <!-- … --> prompt in the template is a placeholder, so each one has to be replaced by real content rather than left in place beside it.

The prompts are read out of the template at runtime rather than listed in the check, so one added to the template is checked from then on with no code change.

Fix: Edit the PR description in the GitHub UI so all four sections are present and no template comment remains.

Reproduce it locally against a description saved to a file — the raw template is itself a usable input:

NODE_OPTIONS='' node --import @swc-node/register/esm-register \
  tools/validation/src/main.ts pull-request-body <path-to-the-body>

🏷️ Ensure Pull Request Labels

Command: pull-request-labels write, the pull-request-labels synchronizer

Runs only on opened/reopened, with continue-on-error: true, so it never fails the job by itself. It reconciles the repository's type:*, scope:*, do-not-merge, source:agent, and source:human labels against configuration/conventional.config.cjs — creating or updating whichever ones drifted. It never deletes: a stale label is reported with the gh label delete command that would remove it, for a human to run.

Fix: A ⚠️ Unable to reconcile labels warning here (for example on a fork pull request without issues: write) does not block the job — the next step still runs and reports its own failure if a label it needs is missing. Reproduce it locally, without mutating anything, with nx run synchronization:pull-request-labels:check.

On a pull request from a fork, GITHUB_TOKEN is read-only no matter what the permissions: block asks for, so this step can only warn and an outside contributor can neither create labels nor label or assign themselves in this repository. Someone with write access here has to add the labels and an assignee on their behalf before 🧾 Validate Pull Request Metadata can go green.

🧾 Validate Pull Request Metadata

Command: validation pull-request-metadata, the pull-request-metadata check

Checks that labels and assignees agree with the title: exactly one type:* label equal to the title's type, exactly the scope:* labels named by the title's scopes (commitlint allows several, split on , or /), no do-not-merge label, at least one assignee, and exactly one source:* label (source:agent or source:human) declaring who opened the pull request.

Fix, by failure mode:

• Missing or mismatched type label — ❌ Expected exactly one type label: type:feat (found: none) — gh pr edit <number> --add-label type:feat, removing any extra type label first.
• Missing scope label — ❌ Missing scope label: scope:callidescope — gh pr edit <number> --add-label scope:callidescope.
• Unexpected scope label — ❌ Unexpected scope label: scope:tools — gh pr edit <number> --remove-label scope:tools.
• No scope in the title at all, such as chore: 🔧 tidy the workspace — commitlint's scope-empty rule rejects this in 📝 Validate Pull Request Title, so in a workflow run this step is never reached; running the check by hand still reports ❌ No scope in title: retitle as chore(<scope>): … — retitle the pull request.
• do-not-merge label present — ❌ Blocked by the do-not-merge label — gh pr edit <number> --remove-label do-not-merge.
• No assignee — ❌ No assignee — gh pr edit <number> --add-assignee @me.
• Missing, extra, or duplicate source label — ❌ Expected exactly one source label: source:agent or source:human (found: none) — remove any stray source:* label and add exactly one of gh pr edit <number> --add-label source:agent or gh pr edit <number> --add-label source:human.

Every failure line in the step output comes with its own gh pr edit remediation command — run the printed commands rather than retyping them.

Reproduce it locally against a real pull request, reading only:

NODE_OPTIONS='' node --import @swc-node/register/esm-register \
  tools/validation/src/main.ts pull-request-metadata <number>

📈 Validate Pull Request Release Significance

Command: validation pull-request-release-significance, the pull-request-release-significance check

This repository squash-merges with PR_TITLE, so the title is the only thing semantic-release ever sees — every commit is discarded once squashed. This check reads the pull request's own commits through gh pr view --json title,commits and fails when the title's type ranks less release-significant than its most significant commit under release.config.cjs's releaseRules, or when a commit uses a scope the title's scopes do not name. See commit-code's Release Significance section for the full type-to-bump mapping.

Fix, by failure mode:

• Title less significant than a commit — ❌ Commit 44dd0cc (feat(validation): check the pull request metadata) needs a title of at least minor significance (e.g. feat): retitle with a more significant type — retitle the pull request to the named type (or higher), or move the offending commit to its own branch.
• Commit scope missing from the title — ❌ Scope "synchronization" is used in commit bbb2222 but missing from the title — retitle the pull request to include that scope alongside the ones already there.

Neither failure mode carries a gh pr edit remediation command — a corrected title changes the subject too, which nothing can infer, so retitle by hand in the GitHub UI.

Reproduce it locally against a real pull request, reading only:

NODE_OPTIONS='' node --import @swc-node/register/esm-register \
  tools/validation/src/main.ts pull-request-release-significance <number>

👮 Audit Issues

Not a pull request check. It fires on issues events, so it never appears among a pull request's checks and never blocks a merge — the issue already exists by the time it runs. It surfaces as the README's Audit Issues badge going red, or as a failed run on the workflow's own page.

🏷️ Reconcile Issue Labels

Command: synchronization issue-labels, the issue-labels writer

Runs only on opened, and always exits 0 — a label it could not add is a fact about the environment rather than a defect — so this step never fails the job. A missing label shows up as the metadata check below failing instead.

🧾 Audit Issue Metadata

Command: validation issue-metadata, the issue-metadata check

Fails when an issue does not carry exactly one type:* label, at least one scope:* label, and exactly one source:* label — or when those labels disagree with the Type and Scope answers in its own issue.yml body. An issue filed through gh issue create has no form markers, so only the presence rules apply to it.

Every failure is reported with the gh issue edit command that fixes it. Fix the labels on the issue itself; no branch change can clear this, and a re-run replays the same issue.

Reproduce it locally against a real issue, reading only:

NODE_OPTIONS='' node --import @swc-node/register/esm-register \
  tools/validation/src/main.ts issue-metadata <number>

🕵️ Scan Security

Each step runs independently:

🚰 Gitleaks Check — nx run codebase:gitleaks --configuration=ci

Config: configuration/gitleaks.toml

Fix: Remove the detected secret from the source file. If it is a false positive, add an [allowlist] entry to gitleaks.toml with a targeted regexes or paths rule.

Never commit credentials — rewrite history if necessary.

🥷 Bandit Security Scan — nx affected --target=bandit

Config: pyproject.toml ([tool.bandit])

Fix: Address the reported security issue (e.g., use secrets.token_hex() instead of random, parameterize SQL queries). Use # nosec <code> only when justified with a comment.

📚 Dependency Audit — part of nx affected --target=scan-security

Fix: Upgrade the vulnerable dependency to a patched version, or add a pnpm audit --ignore entry if no fix is available and the risk is accepted.

🏗 Trivy Infrastructure Scan

Config: configuration/trivyignore

Scan target: infrastructure/terraform/

Fix: Address the CRITICAL/HIGH finding in the Terraform config, or add a scoped entry to trivyignore with a justification comment.

🧑‍🔧 Make Codebase

Triggered only when .devcontainer/** files change (or on manual dispatch). Each step is independent:

🧩 Check VSCode Extensions — nx run codebase:sync-vscode-extensions:check

Config: scripts/sync-vscode-extensions.ts

Source: .vscode/extensions.json, .devcontainer/local/devcontainer.json

Fix: Run npx nx run codebase:sync-vscode-extensions and commit the generated changes.

🔧 Docker Build — devcontainers/ci@v0.3 (Make Codebase step)

Config: .devcontainer/cloud/devcontainer.json

Image: ghcr.io/jimmypaolini/codebase-devcontainer (pushed to GHCR on main only)

Common failures:

• Dockerfile syntax error: Check the Dockerfile referenced in devcontainer.json
• Missing dependency: Add the package to the Dockerfile RUN layer
• GHCR auth failure: packages: write permission is required — check workflow permissions
• Build cache miss causing timeout: Increase runner resources or optimize layer order

🔬 Test Devcontainer — bash .devcontainer/scripts/test-devcontainer.sh

Script: .devcontainer/scripts/test-devcontainer.sh

Fix: Read the test script, identify the failing assertion, and fix the devcontainer configuration or Dockerfile.

Step 4: Apply the Fix

1. Read the relevant source and config files before editing.
2. Apply the smallest change that resolves the error.
3. Follow AGENTS.md conventions (strict TypeScript, explicit return types, no any, consistent-type-imports).
4. Do not commit or push — hand back to the user for review.

Step 5: Verify Locally

Run the equivalent Nx target before handing back:

# Analyze code
pnpm exec nx affected -t lint-codebase

# Test coverage
pnpm exec nx affected -t vitest --configuration=coverage --parallel=3

# Validate conventions (config sync checks only)
npx nx run synchronization:conventional-config:check
npx nx run synchronization:pull-request-template:check
npx nx run synchronization:skill-exclusions:check

# Security
pnpm exec nx run codebase:gitleaks
pnpm exec nx affected -t scan-security

Root Cause & Prevention

You are in triage mode because a proactive validation step was skipped before the commit or push.

After resolving these CI failures, remind the user: use the validate-code skill after every implementation task to catch these issues locally before they reach CI.

Specifically, after every implementation task:

# Auto-fix format, lint, and unused-code issues
pnpm exec nx affected --target=lint-codebase --configuration=write --base=main

# Verify all checks pass — do not push until this is clean
pnpm exec nx affected --target=lint-codebase --configuration=check --base=main

Running this loop locally catches 100% of lint-codebase CI failures — typecheck, lint, format, spell-check, unused code, and sync checks — without waiting for CI to report them.

Step 6: Report Errors Found and Fixes Implemented

The skill ends here. Do NOT do anything else.

At the end of the run, report a summary to the user covering:

1. Errors found — for each failing workflow/job/step, state:

   • Which workflow and job/step failed
   • The specific error message or rule violation that was triaged

2. Fixes implemented — for each fix, state:

   • Whether it was an auto-fix command or a manual code/configuration edit
   • Which files were modified

3. Validation — for each fix, state:

   • The command(s) run to validate the fix
   • The pass/fail result for each command

4. Remaining actions — list any unresolved items requiring user action, or explicitly state all failures are resolved.

Use this report template:

Errors Found
- <workflow>: <job/step> — <error message>

Fixes Implemented
- <auto-fix command and/or manual change>
- Files changed: <file list>

Validation
- <check command> — ✅ PASSED

Remaining Actions
- <none | explicit follow-up actions>