The Agent Skills Directory

[SAFE]: The skill implements a monitoring workflow that delegates logic to local Node.js scripts (wake.mjs and goal-rearm.mjs), ensuring that job status evaluation follows a fixed programmatic ladder rather than potentially exploitable natural language reasoning.- [SAFE]: External inputs, such as "pause declarations," are processed using strict regular expression parsing and can be cryptographically verified using HMAC or Ed25519 signatures, which mitigates the risk of unauthorized status manipulation.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted job signals and prose descriptions entered as data. The risk of injection is minimized by the use of JSON boundary markers, explicit instructions to treat declarations as data, and robust sanitization via regex and signature verification.- [COMMAND_EXECUTION]: The skill invokes the GitHub CLI (gh) and Node.js to perform monitoring and state management tasks. These commands are executed locally and are restricted to checking the state of specific issue/PR identifiers via a well-known service API.

re0-watch