The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and process untrusted repository data, including source code and existing documentation files.
Ingestion points: The Workflow and When a flow doc is required sections in SKILL.md and references/flow-doc/workflow.md instruct the agent to read AGENTS.md, source code files, and existing docs/flows/ markdown files.
Boundary markers: No specific delimiters or "ignore instructions" markers are requested for the ingested content.
Capability inventory: The skill has capabilities to write files, execute the bundled Python validator script, and run repository tools like git and pnpm.
Sanitization: The provided validator script (scripts/validate_flow_doc.py) performs structural validation using regular expressions but does not include sanitization against injection attacks; however, the skill's primary purpose is documentation maintenance rather than executing ingested data.
[COMMAND_EXECUTION]: The skill involves the execution of standard development environment commands and a bundled script.
Evidence: The Workflow and Revision and validation sections mention running python3 to execute scripts/validate_flow_doc.py, as well as pnpm docs:check and git commands.
The bundled validator script utilizes only the Python standard library to perform file reading and string matching, showing no evidence of dangerous operations like network access or arbitrary code execution.

local-dev