hunt-run — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited hunt-run (Agent Skill) and scored it 45/100 (orange). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 4 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 4 flagged
A base64 string of 128+ characters appears in a documentation file. Encoded prompt injection hides the hostile instruction in base64 — invisible to keyword filters — and relies on the agent's ability to decode it at runtime. There is no normal authoring reason to embed a multi-hundred-byte base64 blob in skill docs.
*.sig, SIGNATURES) outside the documentation.A base64 string of 128+ characters appears in a documentation file. Encoded prompt injection hides the hostile instruction in base64 — invisible to keyword filters — and relies on the agent's ability to decode it at runtime. There is no normal authoring reason to embed a multi-hundred-byte base64 blob in skill docs.
*.sig, SIGNATURES) outside the documentation.A base64 string of 128+ characters appears in a documentation file. Encoded prompt injection hides the hostile instruction in base64 — invisible to keyword filters — and relies on the agent's ability to decode it at runtime. There is no normal authoring reason to embed a multi-hundred-byte base64 blob in skill docs.
*.sig, SIGNATURES) outside the documentation.A base64 string of 128+ characters appears in a documentation file. Encoded prompt injection hides the hostile instruction in base64 — invisible to keyword filters — and relies on the agent's ability to decode it at runtime. There is no normal authoring reason to embed a multi-hundred-byte base64 blob in skill docs.
*.sig, SIGNATURES) outside the documentation.Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
<objective> Execute a hunt phase.
Documented flags are available behaviors, not implied active behaviors. Treat --wave N, --gaps-only, and --interactive as active only when the literal token appears in $ARGUMENTS. If none of these tokens appear, run the standard full-phase execution flow.
The execution boundary is the shared THRUNT runtime contract:
QuerySpecthrunt-tools runtime list-connectors, thrunt-tools runtime doctor, thrunt-tools runtime smoke, thrunt-tools pack render-targets, and thrunt-tools runtime execute--wave N executes only a single wave and must not mark the whole phase complete until no incomplete plans remain.
Creates or updates:
.planning/QUERIES/*.md.planning/RECEIPTS/*.mdSUMMARY.md.planning/STATE.md.planning/HYPOTHESES.md when new pivots emerge.planning/HUNTMAP.md when phase status changesAfter this command: Run /hunt-validate-findings <phase>. </objective>
<context> Available optional flags (documentation only):
--wave N — run only Wave N--gaps-only — run only gap-closure plans--interactive — stop after each wave for operator reviewActive flags must be derived from $ARGUMENTS. Do not infer that a flag is active just because it is documented in this prompt. --interactive is active only if the literal --interactive token is present in $ARGUMENTS. If none of these tokens appear, run the standard full-phase execution flow. </context>
<execution_context> @.github/thrunt-god/workflows/hunt-run.md @.github/thrunt-god/templates/query-log.md @.github/thrunt-god/templates/receipt.md @.github/thrunt-god/templates/summary-standard.md </execution_context>
<process> Execute the hunt run workflow from @.github/thrunt-god/workflows/hunt-run.md. Every non-trivial claim must cite receipts. Parallelize by telemetry domain when it helps. When query execution occurs, treat /hunt-run as a QuerySpec producer and normalized-result consumer. If the requested phase has not been planned yet, stop and instruct the operator to run /hunt-plan <phase> first instead of improvising execution. Keep query-log related_receipts and receipt related_queries links exact and bidirectional for artifacts created in the run. Before closing out, update HYPOTHESES.md, STATE.md, and HUNTMAP.md so hypothesis confidence and phase completion match the receipts actually collected. When updating HUNTMAP.md, sync all affected surfaces: phase checkbox, per-plan checklist entries, and the progress table row for the executed phase. When onboarding or debugging a real connector, use thrunt-tools runtime doctor [<connector-id>] before running hunts, and use thrunt-tools runtime smoke [<connector-id>] for a live read-only certification query. When a phase is explicitly pack-backed, prefer thrunt-tools runtime execute --pack <id> or inspect the generated specs with thrunt-tools pack render-targets <id> before running. </process>
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.