The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions in verification-before-completion/SKILL.md employ coercive threats, stating 'If you lie, you'll be replaced', which is a pattern used to override an AI agent's internal safety and honesty protocols. It also uses 'Iron Laws' and mandates that 'violating the letter... is violating the spirit' to force compliance.
[INDIRECT_PROMPT_INJECTION]: The systematic debugging framework requires the agent to ingest and act upon untrusted external data sources.
Ingestion points: The agent is instructed to read error messages, stack traces, and git diffs (SKILL.md files).
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat external log data as untrusted.
Capability inventory: The skill includes a shell script (find-polluter.sh) that executes npm test and provides instructions for using execFileAsync and codesign.
Sanitization: There is no requirement for the agent to sanitize or escape data extracted from logs or diffs before processing.
[CREDENTIALS_UNSAFE]: Diagnostic logging examples in systematic-debugging/SKILL.md instruct the agent to output sensitive information such as the IDENTITY environment variable and keychain identities via the security command.
[COMMAND_EXECUTION]: The provided find-polluter.sh script executes shell commands and runs tests via npm test based on user-supplied search patterns, which could lead to arbitrary code execution if pointed at malicious test files.

debugging