The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute standard development tools including git for branch management, typecheckers, and test suites. It also provides guidelines for constructing diagnostic feedback loops using tools like Playwright or Puppeteer. These operations are intended for local development environments and are limited to authorized project scopes.
[INDIRECT_PROMPT_INJECTION]: The skill processes external inputs such as 'originating agreements', 'tasks', and 'specifications' from potentially remote trackers. This ingestion point allows external content to enter the agent's context. Evidence chain: (1) Ingestion points include task tracker configurations and specification files defined in references/project-documents.md. (2) Boundary markers are implemented through 'context pointers' to authoritative records. (3) Capability inventory includes git merging, subagent spawning, and subprocess execution for verification. (4) Sanitization is addressed via a specific policy in references/diagnose.md requiring the redaction of secrets and credentials from all outputs.
[DATA_EXFILTRATION]: The skill identifies and mitigates risks associated with sensitive data handling. Specifically, references/diagnose.md requires agents to redact secrets (e.g., auth headers, credentials) and replace them with <REDACTED> before sharing any command output or captured artifacts. It also emphasizes keeping credentials in environment variables rather than command strings.

divide-and-conquer