Global

Diagnose and fix failing CI checks by classifying the failure into a known pattern and applying a codified remediation. Use when CI reports MSVC C4996 warnings-as-errors, CMake FetchContent shallow-clone failures, __cpp_lib_format probe mismatches, or when pr-work escalates a failing workflow. Cuts the push-wait-fail-retry loop to a single iteration for the three dominant patterns documented from prior sessions.

Design and build domain-specific agent team harnesses. Analyzes project domains, selects architecture patterns (Pipeline, Fan-out/Fan-in, Expert Pool, Producer-Reviewer, Supervisor, Hierarchical, Generator-Evaluator, Long-Running Session), defines specialized agents (.claude/agents/), and generates skills (.claude/skills/) with orchestration. Use when building new agent teams, designing multi-agent workflows, creating domain harnesses, restructuring agent architectures, setting up long-running multi-session projects, or configuring agent frontmatter (tools, memory, permissions).

Analyze and fix failed CI/CD workflows for a pull request. Use when CI checks fail, GitHub Actions show red, build/test/lint errors block a PR, or the user says 'fix CI', 'fix the build', 'PR is failing', or 'check failed'. Supports solo, team, and batch modes with automated retry and escalation.

Create a release with automated changelog generation from commits since last release using semantic versioning.

Conduct structured research on any topic: web search, codebase analysis, and document synthesis into organized reports. Use when investigating technologies, analyzing alternatives, gathering reference materials, fact-checking claims, or producing technical documentation from research. Use this skill whenever the user asks to research, investigate, compare, or survey a topic.

Generate document index files (manifest, bundles, graph, router) for project documentation. Creates docs/.index/ with searchable registry, feature-grouped bundles, cross-reference dependency graph, and query routing. Supports flat mode (generic projects, script-driven) and grouped mode (projects with doc_id frontmatter, Claude-curated).

Reproduce CI checks locally before pushing so failures surface on the developer machine, not on GitHub. Aligns with the ci-fix skill's pattern catalogue and catches MSVC C4996, CMake FetchContent shallow-clone, and deprecated-API issues pre-push. Invoke manually with /preflight or automatically via CLAUDE_PREFLIGHT=1 in the pre-push hook.

Parse sonarcloud[bot] PR comments, classify findings, codify whitelisted auto-fixes, escalate the rest.

Comprehensive markdown document review - anchors, accuracy, SSOT, cross-references, and redundancy analysis.

Automate GitHub issue workflow - select issue, create branch, implement, build, test, and create PR.

Create well-structured GitHub issues using the 5W1H framework with proper labels and acceptance criteria.

Manage Hazard and Risk records for projects on the regulated-industry track. Maintains a single normalized risk file (docs/.index/risk-file.yaml) holding hazard identification, initial and residual risk estimates, control measures with verification links, and bidirectional Risk<->Requirement linking via the requirements[] field. Subcommands: add | edit | evaluate | validate | list. Output is consumed by the traceability skill (matrix risk_ids[] field) and the evidence-pack skill (risk_file kind). Opt-in: no-op when docs/.index/manifest.yaml is absent so non-regulated repos are unaffected. Atomic writes via *.tmp + rename; idempotent for diffability. Implements ISO 14971 sections 5-9 operationally.

Generate and validate the bidirectional traceability matrix linking requirements, design, code, tests, risk records, and standard clauses. Consumes docs/.index/{manifest,bundles,graph,router}.yaml plus an optional compliance/ directory and produces docs/.index/traceability.yaml (machine-readable) and docs/.index/traceability.md (human-readable). Read-mostly: writes only the two trace artifacts and never mutates source documents. Opt-in — no-op when docs/.index/graph.yaml is absent so non-regulated repos are unaffected.

Assemble per-release evidence packages from artifacts already produced elsewhere in the project (traceability matrix, CI logs, PR review records, signed-commit lists, external research citations, risk file). Output is a self-contained directory under evidence/<version>/ containing a manifest.yaml, per-artifact subdirectories, and an INTEGRITY checksum. Read-mostly with respect to source code: writes only under evidence/. Opt-in gate: no-op when the consumer project has no evidence/ parent directory, so non-regulated repos are unaffected. Idempotent: refuses to overwrite an existing pack unless --force is passed.

Interactive review of stale, flagged, and duplicate memories from the latest audit report. Walks entries one at a time with verify / quarantine / edit / skip choices.

Clean up merged and stale branches from local and remote repositories. Use when branch list is cluttered or after merging PRs.

Maintain a SOUP (Software Of Unknown Provenance) register for every third-party software item the project depends on. Discovers candidates from lockfiles (package-lock.json, go.sum, Cargo.lock, requirements.txt, pyproject.toml, pom.xml, packages.lock.json), enriches with human-supplied risk class and verification refs, validates against a license allow-list and the requirements catalogue, and emits a per-supplier report. Outputs docs/.index/soup.yaml plus docs/.index/soup.md. Subcommands: discover | enrich | validate | list | report. Bidirectional linking with traceability via the soup_ids[] field on requirement rows. Opt-in: no-op when no lockfile is detected and docs/.index/soup.yaml is absent. Atomic writes (*.tmp + rename); idempotent (records sorted by id). Implements IEC 62304 sections 5.3.3 (SOUP requirements) and 8.1.1 (configuration items).

Fan out a single high-level directive (audit, deprecation cleanup, fix, migration, version bump) across an arbitrary list of repositories as parallel Agent workers, with a supervisor that polls a shared manifest, renders a live status table, and aggregates final results. Use when the user says 'apply X across repos', 'audit N repositories', 'sweep every repo', 'run the same fix in all projects', 'parallel multi-repo', 'fleet-wide change', or provides a list of repositories with a single directive. Preferred over running issue-work in each repo sequentially.

Enforce complete implementation of all tiers/difficulty levels for tiered features. Prevents partial implementations.