The Agent Skills Directory

[DATA_EXPOSURE]: The skill accesses the sensitive configuration file ~/.claude.json located in the user's home directory to detect configured MCP servers.
Evidence: Found in commands/claim.md and commands/spawn.md where the skill reads the file to find Linear MCP instances.
Context: This file often contains plain-text API tokens (such as LINEAR_ACCESS_TOKEN) and environment variables for other MCP services. While used for integration, this represents access to high-value secrets.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external platforms which could influence the agent's behavior.
Ingestion points: Fetches PR comments via the GitHub CLI (peep.md) and ticket descriptions from Linear (cook.md).
Boundary markers: The skill does not implement explicit delimiters or warnings to ignore instructions embedded in the external text.
Capability inventory: The skill possesses extensive capabilities including file system modification, command execution via pnpm and npx, and PR management via gh.
Sanitization: External content from comments and tickets is interpolated directly into prompts for planning and code generation without sanitization.
[COMMAND_EXECUTION]: The skill frequently executes system-level commands and development utilities.
Evidence: Usage of git, gh, pnpm, npx, lsof, and kill across multiple command workflows.
Context: Commands are used for standard development tasks such as scaffolding projects (pnpm create), managing branches, and running diagnostics (e.g., npx -y react-doctor@latest).

bruhs