The Agent Skills Directory

[SAFE]: The skill defines a comprehensive security architecture designed to prevent unauthorized actions. It utilizes a deterministic JSON-based 'autonomy ledger' (enforced by scripts/check_autonomy_ledger.py) to gate any autonomous external mutations like PR creation or Jira transitions, requiring explicit user-bound approval hashes.
[COMMAND_EXECUTION]: The file scripts/test_autonomy_ledger.py contains calls to subprocess.run. Analysis shows these are used exclusively to execute a local Python validation script against test fixtures during the development/testing phase. The command construction uses absolute path resolution and the current system interpreter, representing a safe and standard testing pattern.
[INDIRECT_PROMPT_INJECTION]: The skill manages a surface for indirect prompt injection by processing external initiative documentation and contracts. However, it implements strong mitigations, including 'Fast Contracts' (detailed in references/fast-contract.md) that provide a strict boundary for sub-agents, and multiple mandatory human-in-the-loop and automated review gates (Document Review, Code Review, Security Sentinel) to validate all generated artifacts.
[SAFE]: Instructions throughout the reference files (e.g., references/safety.md, references/release-handoff.md) explicitly forbid the harvesting of credentials, advising the agent to use deterministic checkers instead of asking for secrets, and ensuring that 'worker' roles are restricted from performing shipping operations.

krt-compound-master