The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands locally using pnpm, including pnpm run lint, pnpm run fmt, and pnpm exec commitlint. These are standard development operations but grant the agent capability to run local subprocesses.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks because it processes untrusted commit message data from pull requests.
Ingestion points: Commit messages are ingested into the agent context via the $MSG variable used in SKILL.md.
Boundary markers: No boundary markers or specific "ignore instructions" directives are present to separate the untrusted commit message content from the shell command template.
Capability inventory: The agent has the capability to execute arbitrary shell commands via the pnpm workspace tools as defined in SKILL.md.
Sanitization: There is no evidence of sanitization, escaping, or validation of the commit message string before it is interpolated into the shell command printf '%s' "$MSG" | pnpm exec commitlint.

code-review