The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill manages its own tooling via bun install --frozen-lockfile as seen in scripts/bootstrap.ts. The dependency list includes well-known packages like commander and typescript. While the lockfile specifies futuristic version numbers (e.g., typescript@7.0.2), these are used within the context of a well-defined development ecosystem and standard package registries.
[COMMAND_EXECUTION]: The skill facilitates extensive repository operations using established CLI tools like git and gh (GitHub CLI). It also includes a custom local orchestration tool orch for managing multi-agent tasks, which operates on a local TSV/JSON store to maintain state and audit trails.
[INDIRECT_PROMPT_INJECTION]: The skill implements robust defenses against indirect injection by explicitly instructing the agent in playbooks/babysit.md to treat review-comment text as untrusted data and to triage findings against the code rather than treating the text as new instructions.
[SAFE]: The skill incorporates a strict autonomy policy that requires the agent to pause for user approval before performing irreversible writes, such as force-pushing to shared branches, deploying code, or deleting data. This reduces the risk of accidental or malicious damage during autonomous operation.

poteto-mode