The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and validate untrusted external data in the form of GitHub Pull Requests, commit histories, and build logs.
Ingestion points: Untrusted data enters the agent context via Git checkouts, gh api repository metadata, PR descriptions, and build output (captured in SKILL.md and references/patch-refresh.md).
Boundary markers: The instructions mandate the use of a "private worksheet" (assets/validation-worksheet.md) to track evidence and explicitly require humans to judge interactive UX and review publication drafts.
Capability inventory: The skill utilizes git, gh, node, and environment-specific build tools (Node.js, Python, etc.), as well as specialized environment managers (ocm). It can also interact with Docker if authorized (references/diagnostics.md).
Sanitization: The scripts/report.mjs helper includes HTML escaping for generated reports. The skill instructions warn against including raw logs, telemetry, or credentials in public comments and require human privacy review.
[COMMAND_EXECUTION]: The skill utilizes several helper scripts that execute system commands to facilitate its workflow.
Evidence: scripts/check-update.mjs uses execFileSync to invoke git for repository state checks and gh to query the GitHub API. scripts/report.mjs reads and parses JSON files to generate Markdown summaries.
Context: These executions are restricted to checking the status of the skill's own source repository and processing validation reports, which is consistent with its primary purpose.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to check for updates to its own source code.
Evidence: scripts/check-update.mjs queries the GitHub API (api.github.com) to retrieve commit information from the official vendor repository TheAngryPit/TheAngrySkills.
Context: This behavior targets a well-known service (GitHub) and a repository belonging to the skill's author to ensure the environment is running the expected version.

openclaw-custom-build-validation