The Agent Skills Directory

[SAFE]: No malicious patterns detected. The skill performs project management automation within a defined workspace.
[COMMAND_EXECUTION]: Uses dynamic context injection to verify the execution environment via !echo "${BMAD_AUTO_MODE:-}". This is a benign check for environment variables used to gate execution.
[DATA_EXPOSURE]: Accesses project-specific files like config.yaml and deferred-work.md to perform its primary function. No sensitive system paths (e.g., credentials or SSH keys) are targeted.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources (the ledger file and codebase) to verify task statuses. This represents an attack surface where instructions embedded in code or the ledger could influence the agent.
Ingestion points: {implementation_artifacts}/deferred-work.md and source code files.
Boundary markers: Not explicitly defined for the ingested data content.
Capability inventory: File read, file write (JSON result/ledger migration), and sub-agent invocation.
Sanitization: No specific sanitization or filtering of the ledger content is mentioned.
However, the skill's impact is restricted by its machine-readable output format and limited write access, resulting in a low risk profile.

bmad-auto-sweep