Skip to content

Skill providers

Who else provides what each HQ skill does, so a skill is retired when someone else does it as well or better (HQ as a product, Phase 1). The choice per capability is recorded in [capabilities] in repos.toml; this file holds the evidence behind it. Surveyed 2026-11-19 from vendors’ own repos and docs, skill marketplaces and GitHub’s agentic workflow (gh-aw) catalogues.

How sure: catalogue listings were checked against the source repo where it mattered (the OSPO workflows, the gh-aw engine and cross-repo docs, the security tools’ release dates). Star counts and marketplace entries were not. Several marketplaces (skills.sh, claudemarketplaces.com and similar) list skills with no review, so an entry there isn’t a recommendation.

  • No third-party provider covers an HQ capability fully. The closest are GitHub’s OSPO workflows (gh-aw), on trial from 2026-11-19 beside four HQ skills (below).
  • Nothing exists for forum watching, reputation, alignment across repos, agent accessibility, adoption trends across package registries, roadmaps, policies or projects. These stay HQ skills.
  • Platform knowledge for Home Assistant and SignalK exists only as third-party skills of unknown upkeep. None is official. Not adopted; worth a look when the architecture review next runs.
  • Security is where reuse pays most: deterministic scanners feed one place (code scanning), and the HQ skill becomes the step that merges and ranks them.

Overlap: full (could replace the skill), partial (covers part; HQ keeps the rest), none.

Capability (HQ skill) Best candidates Maintainer, kind Overlap Decision
support (support-sweep) OSPO Organization Health Report; issue-triage GitHub, gh-aw partial: stale issues and “needs a human” items, but one weekly issue, not a ranked daily list of people waiting trial
pull-requests (pr-report) OSPO Organization Health Report (stale PRs, merge times); Contribution Check GitHub, gh-aw partial: no split of external contributions, stuck bumps and drafts trial (same run as support)
dependency-updates (dependabot-sweep) dependabot skill; Renovate automerge GitHub, skill; Mend, tool partial: config help only. Notes that @dependabot merge comment commands were deprecated (Jan 2026) keep; the shared auto-merge workflow already does the deterministic part
security (security-escalation) see Security partial merge step over several providers
registry (registry-check) OSPO Stale Repository Report GitHub, gh-aw partial: inactive repos only, no comparison with repos.toml or publishing channels trial
github-settings (github-review) OSS Release Compliance Checker; CI Doctor and CI Coach (gh-aw); OpenSSF Scorecard GitHub, gh-aw; OpenSSF partial: community files and security settings per repo; nothing compares workflows across repos trial (compliance checker); Scorecard already in use
adoption (adoption-report) OSPO Contributors Report GitHub, gh-aw partial: contributors only; no provider reads PyPI, npm, GHCR, HACS or HA analytics trial for the contributor part; downloads are HQ Stats
docs (docs-review) Markdown Linter, Link Checker, Documentation Unbloat, Daily Accessibility Review (gh-aw) GitHub Next, gh-aw partial: each covers one aspect, none checks fidelity to the code keep; revisit a link checker as a cheap add-on
releases (release-digest) Daily Repository Chronicle (gh-aw); changelog skills on marketplaces GitHub Next; third parties partial: no judgement of what’s significant to users keep
support-themes (support-review) Weekly Issue Activity, Discussion Task Miner (gh-aw) GitHub Next, gh-aw partial: issues only, no forums keep
architecture (architecture-review) Home Assistant quality-scale skills on marketplaces third parties, unverified partial: quality scale only keep; check upkeep before any use
platforms (platform-review) Home Assistant and SignalK development skills on marketplaces third parties, unverified none for release watching keep
roadmaps, projects gh-aw plan and team-status workflows GitHub, gh-aw partial: planning issues, not review against platform change keep
alignment, agent-accessibility, forums, reputation, policies none found none keep

Four of GitHub’s OSPO workflows run in hq-private from 2026-11-19 for two weeks, beside the HQ skills they overlap. Each writes one issue labelled trial (never assigned, so no one is paged).

  • Engine: Claude (engine: claude, an ANTHROPIC_API_KEY secret). gh-aw doesn’t accept subscription tokens. Copilot-hosted inference is the alternative and needs no secret, only a Copilot plan.
  • Reach: gh-aw reads every public repo by default, so running from one repo covers the org. Private repos would need a token, and they’re out of scope anyway.
  • Pinned to awesome-copilot commit 1683f50. The upstream instructions are unchanged. Local settings are the engine, the schedule (weekly for the trial), the trial label and a marked section at the end of each file. The compliance checker runs only on issues labelled ospo-release-check, not on every new issue.
  • Question (reframed 2026-11-19): is a workflow worth supporting as an optional HQ provider, off by default and turned on per instance like a skill, perhaps as a quarterly audit? The HQ skills stay either way; nothing is retired on the trial’s result. The default HQ must stay low-friction for a developer with a handful of projects, so set-up cost counts as much as output quality.
  • Compare on the same weeks: what each found that the other missed, false positives, whether the output needs a human to act, how much set-up and tuning it needed, and the cost per run. Results go in the project log.
  • Findings so far: the first org health run (2026-11-19) cost about $2.43 on Sonnet 5 and filed nothing. Told to cover public repos only, it searched repo by repo, hit GitHub’s search rate limit and ended without its safe output. The local settings now keep searches org-wide with is:public, one at a time, and always file the issue (hq-private#129). The re-run the same evening cost about $1.09 and filed hq-private#130: an accurate snapshot (activity, merge times, four stale issues) that says which figures it couldn’t get (commit search blocked in its sandbox, PRs sampled at 100), but with nothing to act on. It has no notion of someone waiting on the maintainer, counts the maintainer’s own roadmap items as stale, and its thresholds suit a large org (a 94% merge rate shows yellow). So far it fits as an optional periodic snapshot, not a replacement for support-sweep or pr-report.

Already running beside Dependabot alerts and CodeQL: secret scanning with push protection (every public repo), OpenSSF Scorecard (every released repo) and Grype (container images). Everything that uploads SARIF lands in code scanning. One API (code-scanning/alerts, tool.name) then reads all of it, so /security-escalation merges them in one place.

Tool Finds Output Fit
zizmor Actions workflow flaws: template injection, unpinned or impostor actions, excess permissions, credentials left in checkouts SARIF to code scanning chosen. Every repo runs the shared admin workflows, and Scorecard only scores these issues. Shared workflow reusable-zizmor.yml, reaching repos with the next hq release. Free
Socket.dev malicious or risky npm and PyPI packages with no CVE (install scripts, typosquats, new maintainers) GitHub App: PR checks and comments chosen, pending the maintainer (a GitHub App install). Nothing else catches this. Free for open source per its pricing page (not checked against the terms)
Trivy images, filesystem, IaC, secrets SARIF not now: Grype already covers the images
osv-scanner, pip-audit, npm audit known vulnerable dependencies SARIF / JSON not now: a second view of the same advisory data as Dependabot
Semgrep CE, Bandit static analysis SARIF not now: overlaps CodeQL; worth it only for custom rules
gitleaks, TruffleHog secrets in history SARIF not now: secret scanning and push protection are on everywhere
StepSecurity Harden-Runner runtime egress on CI runners dashboard later: CI compromise is a real gap, but it has no code scanning output, so a separate read path
claude-code-security-review AI review of PR diffs PR comments not now: last pushed 2026-02, no releases, and its README says it isn’t hardened against prompt injection, so not for fork PRs
Copilot Autofix fixes for code scanning alerts PR suggestions free on public repos; nothing to install
Snyk, Aikido, Endor Labs, Semgrep platform SCA, SAST, triage own dashboards not now: free tiers capped or per developer, and findings don’t reach code scanning

How /security-escalation merges them is in its skill (“Providers and merging”).