The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill indexes arbitrary markdown files (PRDs and checkpoints) from project directories across the local machine. This provides a surface for indirect prompt injection if those files contain malicious instructions intended to influence the agent's behavior during indexing.
Ingestion points: PRD markdown and checkpoint files in .dev/ and .dev-archive/ directories, scanned via discover.sh and processed by dev-workflow.cjs.
Boundary markers: SKILL.md contains explicit instructions to treat scanned content strictly as untrusted data and ignore any directives found within it.
Capability inventory: The skill uses the Bash tool to execute indexing and discovery scripts that write to ~/.dev-wiki/ and modify ~/.config/dev-dashboard/config.json.
Sanitization: The dev-workflow.cjs script uses js-yaml and gray-matter for parsing. The js-yaml implementation includes specific hardening that disables the execution of dynamic JavaScript functions within YAML tags.
[COMMAND_EXECUTION]: The skill executes a large bundled Node.js script (dev-workflow.cjs) and several bash scripts (discover.sh, validate.sh) to perform discovery and indexing tasks. These scripts have the capability to perform broad filesystem operations, including creating symlinks and writing to configuration files in the user's home directory.

dev-wiki