The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external plan files (e.g., docs/PLAN.md) and merges this content into durable project files like AGENTS.md, ARCHITECTURE.md, and ROADMAP.md. 1. Ingestion points: The skill reads file paths provided via arguments or searches for docs/PLAN.md files. 2. Boundary markers: The skill relies on specific text markers (like UNVERIFIED or CORRECTED) to filter content, but lacks robust delimiters or automated sanitization. 3. Capability inventory: The skill performs file writes to existing project documents, executes shell commands for verification, and spawns subagents to test the standalone utility of the documentation. 4. Sanitization: There is no evidence of automated sanitization or escaping of the ingested content before it is written to files. A malicious plan could embed instructions that influence future agent behavior or trigger unintended command execution during the verification phase.
[COMMAND_EXECUTION]: The skill executes several shell commands to verify the state and metadata of the repository. Evidence: Instructions include running git log --oneline, wc -l, git remote -v, and git config to validate metadata and staleness. While these are primarily for verification, their output is directly used to populate documentation files.

deep-plan-ingest