The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the clawpatch package globally using npm install -g clawpatch or pnpm add -g clawpatch. This involves downloading and executing software from a third-party registry not listed in the trusted vendors list.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands to manage project state, including find, test, ps, and various clawpatch subcommands. These commands interact with the local file system and process tree.
[DYNAMIC_EXECUTION]: The skill provides complex shell scripts for 'Forced Re-Review' that use jq to parse featureId values from JSON files and interpolate them into shell commands. This pattern of generating and executing shell commands based on the contents of local data files represents a dynamic execution surface.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process repository code and semantic feature records for automated review via external providers (Codex, Claude, Pi). This processing of untrusted repository data creates an attack surface where malicious code could attempt to influence the LLM provider's review output or agent behavior.
Ingestion points: .clawpatch/features/*.json, .clawpatch/findings/*.json, and the target repository source code.
Boundary markers: None identified in the provided instructions to delimit code from instructions.
Capability inventory: The skill can execute shell commands, write to the file system (patches, reports, locks), and perform network operations via the configured providers.
Sanitization: No specific sanitization or escaping mechanisms for the ingested code are described.

clawpatch