Index of every page on this site: llms.txt
All skills
Run it yourself

Scans a repo's GitHub Actions workflows for the ten critical CI/CD attack vectors — template injection, fork code executed with privileges (pwn requests), cache poisoning, impostor action SHAs, secrets dumps, GITHUB_ENV hijack, write-token untrusted triggers, credentials in caches/artifacts, unverified remote code execution (curl|bash and mutable fetch-and-run), and dependency install scripts running in a job that holds secrets — reports every finding with a plain-English attacker scenario, plus pass/fail config hygiene checks, and fixes selected findings via per-finding subagents. Deliberately NOT comprehensive — critical exploit-chain checks only (references/why-these-ten.md). Use when: the user asks to audit or review CI/CD or GitHub Actions security posture, "is my CI secure", "audit my CI security", or names ci-secure / /ci-secure. Do NOT trigger for: CI speed, cost, or wall-clock audits ("why is CI slow") — use ci-speedup; CI config best-practices grading ("grade my CI", "CI score") — use ci-score.

Use this Skill: https://skilld.dev/gh/starslingdev/skills/ci-secure

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referencesterminal-summary.md

≈2.2k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Composing the terminal summary

Mechanics for SKILL.md Phase 3. SKILL.md holds the rules (what must be said, and what must never be dressed up as a pass); this file holds the provenance of each line and the exact receipt format. Read it before printing the summary.

Contents


The three-line header block

CI Secure   3 critical findings  ▏2 of 10 vectors hit▕  12 workflows · impostor check ran
  Impostor-SHA check (P14.11): ran — 14 unique pins verified, 0 flagged
  Coverage: complete

The first line is the report's own banner, pre-drawn by report.py inside a fenced block immediately under the provenance table. Copy it verbatim; never redraw, re-count or reformat it. The other two are assembled. There is no fourth HEADER line — SKILL.md Phase 3's mandatory contract lines (the hygiene clause, the ten-row vector receipt, the dormancy note) follow this block and are not optional.

Where each line comes from

Line Where it comes from
CI Secure … Pre-drawn — copy, never compose. grep '^CI Secure' "$REPORT". Rendered inside a fenced block under the provenance table with its counts already computed.
Impostor-SHA check (P14.11): … Assembled from the banner's own impostor-check word (ran / partial / SKIPPED / not recorded — the impostor word ENDS the pre-drawn line, and it is two words in the not recorded state, so read to the end of the line rather than taking the last token) plus the gh_checks["P14.11"] status/detail in $FINDINGS (the pin counts — "14 unique pins verified, 0 flagged", or the UNVERIFIED count on a partial — live there, NOT in any report row). On a run that did NOT fully complete, add the reason from the report's > [!WARNING] gh-checks blockquote (or the ⚠️ P14.11 vector-map row). Do NOT read the pin counts off the vector-map row: when the check ran clean that row is a generic ✅ "no match" like every other clean vector and carries none of them. Always say PARTIAL … NOT a pass / SKIPPED … this check did NOT run verbatim when it did not run. Then read the action_lockfile dict the same $FINDINGS one-liner prints on its second line (the JSON, not the report, is the source): when state is valid and covered_references is above 0, append ; N action reference(s) in M workflow(s) locked by actions.lock are verified by GitHub at run time, not by this check (N = covered_references, M = the length of covered_workflows); when state is invalid, append ; <path> is not a lockfile this scan can read, so no reference is counted as locked; when absent, append nothing.
Coverage: … Assembled from the Coverage ROW of the provenance table at the top of the report — NOT from any sentence under the banner, where nothing of the kind is rendered. The row reads ✅ complete — every workflow file was scanned or ⚠️ **PARTIAL** — not every workflow was fully scanned, and on PARTIAL it does not say what was missed: that lives in the separate > [!WARNING] **Incomplete coverage — …** blockquote further down, and your line must carry it.

P14.24's release-immutability lookup is not a header line. It is recorded in gh_checks["P14.24"] only when the scan reached a GitHub CLI extension pinned to a tag and then run, and it can only remove a finding. The Phase 3 $FINDINGS one-liner prints it on its third line (empty when the key is absent). When that status is skipped or partial, say so in one clause where the summary names the P14.24 finding(s): release-immutability lookup did not run (<reason>), so tag-pinned extensions are reported as mutable, adding ; sign in to gh and re-run to check them when the reason is gh sign-in or --gh-impostor=off. Never call this a missing finding, and never word it as NOT a pass: every tag-pinned finding is still listed.

The extraction commands are in SKILL.md Phase 3. Run them as written; do not re-derive them here. The one to be careful with is the P14.11 vector-map row: the id renders in backticks with no space before it, so a pattern expecting | immediately ahead of the id matches nothing and silently drops the line.

The push-credential rows

The report's hygiene section ends with a ### If a push credential leaks sub-heading holding three hygiene rows (sec.environments.branch-restricted, sec.environments.approval-holds, sec.secrets.environment-scoped). They are hygiene checks like the rest: name a failed one in the hygiene clause, in one short clause each (3 environments accept a branch any push credential can make, 3 environments let the requester approve their own run, any workflow pushed to any branch can read 11 secrets), and an unmeasured one as unmeasured with its reason — never 0 secrets over a list that could not be read. Say what must be done in plain words, as the row does; there is no prompt to hand to another agent. Never claim these rows would have stopped an attack by an admin or org owner; they do not, and the report does not say so either.

When sec.secrets.environment-scoped fails, it leads the hygiene clause. The report then carries a > [!CAUTION] Secrets reachable from any branch callout right under the headline; take the count from it and say the consequence, not the bookkeeping: Hygiene: any workflow pushed to any branch can read 11 secrets, so one leaked push credential exposes them all with no review and no merge; move them into an environment only protected branches can use (one that passes the branch-restriction check). When the callout says "at least", say "at least 11" too. Name secrets only as the report names them (names, never values), and never call this an attack-vector hit: the vector count is unchanged. No callout means the row did not fail; an unmeasured row is still said as unmeasured, never as this sentence.

The vector-map receipt format

Derive each line from the report's "Vector map" rows with links and anchors stripped: NUMBERED 1–10 in the report's row order, one per vector, catalog id included, numbers right-aligned — pad rows 1–9 with a leading space so the periods line up under 10.:

Vector scan — 10 attack vectors checked, 1 hit:
  1. ✅ P14.10 Template Injection in run: Blocks — no match in 3 workflows
  2. 🟥 P14.9 Fork code executed with privileges — 2 sites across 2 workflows → Finding 1
 …
 10. ⚠️ P14.11 Impostor / unreachable SHA — SKIPPED (gh not authenticated); this check did NOT run

Icons: ✅ evaluated-clean; 🟥 a HIT whose group severity is HIGH; 🟧 a HIT whose group severity is MEDIUM, each with its site count (2 sites across 2 workflows) and its → Finding N tag; ⚠️ did-not-run, which keeps its reason and is never promoted to ✅.

Plain text only, in the receipt and in the question text that carries it — no **bold**, no headings, no code fences. The question UI supplies its own emphasis, and bolding the whole block made a close unreadable.

The per-finding verdict lines

After the vector map, print one verdict line per finding, so the user sees how worried to be about each one without opening the report. Take them from the render plan, which carries the report's own order and each group's verdict (the agent's Phase 2.5 judgment, flattened to one line):

SLUG="$(printf '%s' "$(git rev-parse --show-toplevel)" | shasum | cut -c1-12)"; FINDINGS="${TMPDIR:-/tmp}/ci-secure-findings-${SLUG}.json"
<ci-secure>/scripts/report.py --render-plan --in "$FINDINGS"
# [{"pattern": "P14.18", "dormant": false, "verdict": "Real in principle but low in practice. ..."}, ...]

List position is the finding number (index 0 is Finding 1). Print the verdict's first sentence, and add (dormant) for a dormant group:

Verdicts:
  Finding 1 (P14.18): Real in principle but low in practice.
  Finding 2 (P14.10): High: anyone who opens a pull request reaches it today.

An empty verdict means Phase 2.5 did not write one for that group: say Finding N (P14.x): no verdict written, never invent one here. The verdict is the agent's judgment of THIS workflow and may be lower than the catalog severity printed in the vector map; say it as written, without softening or raising it. Plain text only, and inside the question text whenever the next act is a structured question (SKILL.md Phase 3's delivery caveat).

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warningtoday3 checks · Risk SAFE
  • Gen Agent Trust Hubtoday

    The skill 'ci-secure' is a comprehensive security auditing tool for GitHub Actions. It is authored by starslingdev and follows excellent security practices, including centralized data sanitization to prevent Markdown injection and defensive prompt engineering to protect against malicious instructions in audited code. Its use of dynamic loading and subprocess calls is restricted to essential functions for modularity and repository analysis using standard tools like git and gh.

  • Socket4d

    No alerts

  • Snyktoday

    Risk: MEDIUM · 2 issues

Signed by skilld at 9a84115. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 16 hours ago.

Activeupdated last week

README badge

README badge for starslingdev/skills/ci-secure