The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of system commands including git, gh (GitHub CLI), ocm (OpenClaw Manager), and project-specific build scripts to perform validation tasks. These are executed via node:child_process in helper scripts.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data from consumer repositories, including PR code and build instructions. This creates an attack surface for indirect prompt injection, which the skill mitigates through mandated process isolation (Crabbox/OCM) and human review of generated reports.
Ingestion points: Processes repository instructions, build scripts, and PR commits as specified in SKILL.md and references/patch-refresh.md.
Boundary markers: Relies on operator review and environment isolation rather than explicit programmatic prompt delimiters.
Capability inventory: Possesses capabilities to execute shell commands, manage local services via OCM, and interact with the GitHub API via the gh CLI.
Sanitization: Implements sanitization for Markdown exports and requires human verification of publication drafts as described in references/reporting.md.
[EXTERNAL_DOWNLOADS]: The scripts/check-update.mjs utility performs read-only network requests to the vendor's repository (github.com/TheAngryPit/TheAngrySkills) via the GitHub CLI to verify the skill's version status.

openclaw-custom-build-validation