The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from GitHub pull requests, CI job logs, and external images, which constitutes a vulnerability surface for indirect instructions.\n
Ingestion points: The skill ingests data from gh pr view, gh run view, and PNG images downloaded via curl from URLs provided by the posthog:visual-review-runs-snapshots-list tool.\n
Boundary markers: The instructions do not define specific delimiters or include explicit "ignore embedded instructions" warnings for the content found within visual screenshots or CI log text.\n
Capability inventory: The agent has the capability to perform network operations (via posthog tools and curl), write to the local filesystem (via curl to /tmp), and modify repository state (via posthog:visual-review-runs-finalize-create).\n
Sanitization: There is no evidence of content sanitization or filtering for the retrieved logs or images prior to analysis.\n- [EXTERNAL_DOWNLOADS]: The skill downloads images from external S3 buckets to perform visual comparisons between baseline and current renders.\n
Evidence: curl -s -o /tmp/vr-baseline.png \"<baseline_artifact.download_url>\"\n
Context: These downloads are a core component of the visual review process. The URLs are provided by the vendor's own API tools, and the data is stored in a temporary directory for inspection.\n- [COMMAND_EXECUTION]: The skill uses standard developer command-line tools to inspect the state of the repository and its CI runs.\n
Evidence: Usage of gh pr view, gh run view, git diff, and curl.\n
Context: These commands are used for their intended purpose within a developer-centric workflow and do not involve suspicious parameters or privilege escalation.

triaging-visual-review-runs