The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from reporter interactions (bug reports) and external documentation (MDX files) to drive its verification logic.
Ingestion points: Processes reporter-named routes and interaction descriptions during the 'Completion boundary' and 'Regression cases' phases; reads MDX files from 'content/docs' and registry artifacts in 'apps/www/src/registry/'.
Boundary markers: While the skill enforces a structured 'Completion boundary' and 'Evidence' workflow, it does not specify the use of delimiters or 'ignore embedded instructions' warnings when processing external text inputs.
Capability inventory: The skill possesses extensive capabilities, including executing shell commands (pnpm, bun, node, curl, git), managing local processes (lsof), and performing browser automation via Playwright.
Sanitization: There is no evidence of sanitization or filtering of external content before it is used to influence the agent's decision-making or command generation.
[COMMAND_EXECUTION]: The skill provides the agent with the authority to execute a wide variety of shell commands and local scripts essential for its verification tasks.
Evidence: Instructions include running 'pnpm', 'bun', 'node', 'curl', 'git', and 'lsof' commands across several files (e.g., SKILL.md, references/commands.md).
Internal Scripts: The agent is directed to run project-specific scripts such as 'tooling/scripts/proof-worktree.mjs', 'tooling/scripts/type-probe.mjs', '.agents/rules/plate-next/scripts/sync-resources.mjs', and '.agents/pstack/mutate.mjs'.
[DYNAMIC_EXECUTION]: The skill frequently uses local Node.js scripts to perform runtime operations like creating worktrees, probing types, or mutating code for testing.
Evidence: Use of 'node' to execute '.mjs' files in the 'tooling/scripts/' and '.agents/' directories, often passing dynamic arguments derived from the current task context.

verify