The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands for source discovery and runs a bundled Python script (scripts/merge_constraints.py) to handle the deterministic merging, ID assignment, and formatting of extracted constraints. These operations are performed locally using standard system tools.
[INDIRECT_PROMPT_INJECTION]: The skill's primary functionality involves ingesting and extracting content from various documentation files in the project repository (e.g., docs/, adr/, rfcs/). This creates a surface for indirect prompt injection if those files contain malicious instructions intended to manipulate the agent.
Ingestion points: Reads markdown and plain text documents from directories such as docs/, adr/, spikes/, and notes/, as well as standard files like README.md and SECURITY.md.
Boundary markers: The skill does not explicitly instruct subagents to use boundary markers or specific delimiters to isolate the content being read from the agent's instructions.
Capability inventory: Capabilities include reading and writing files within the repository, spawning parallel subagents for processing, executing local scripts, and modifying agent-specific configuration files (AGENTS.md, CLAUDE.md).
Sanitization: Extracted statements and impacts are parsed as strings and rendered into markdown formatting without specific sanitization to prevent the interpretation of embedded instruction sequences.

constraints-extractor