The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from implementation issues and existing source code which could contain malicious instructions. However, the risk is mitigated by a narrow execution scope (one design question), explicit boundary rules for prototypes, and a mandatory human review step.
[COMMAND_EXECUTION]: The skill performs shell commands for Git operations (status, worktree, commit) and project-specific tasks (linting, testing, serving UI). These operations are restricted to local repository context and follow documented repository policies.
[DATA_EXFILTRATION]: Instructions explicitly prohibit unauthorized network operations, including git push or external data transfers. The skill enforces local-first development by default.

ptlam-prototyping