The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to use npx ctx7@latest to fetch documentation. This involves downloading and executing a package from the NPM registry at runtime. The use of the @latest tag makes the execution non-deterministic and susceptible to package takeover or malicious updates from an unverified source.
[COMMAND_EXECUTION]: The skill requires the agent to "execute/simulate each step" of a QA checklist generated from uncommitted code changes and issue descriptions. This involves the autonomous execution of shell commands derived from AI-generated plans based on local file content.
[COMMAND_EXECUTION]: The "AFK contract" explicitly prevents the agent from calling AskUserQuestion and requires it to make decisions autonomously. This removes the primary security control of human-in-the-loop review before executing generated commands or pushing code to remote repositories.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes data from local *.md files to drive its implementation logic and QA simulation.
Ingestion points: Markdown issue files located in the target folder (e.g., docs/issues/).
Boundary markers: None; the agent directly parses and follows instructions within the ## Blocked by and acceptance criteria sections.
Capability inventory: Subprocess execution (tests/verification), file writing, git operations, and gh CLI for PR creation.
Sanitization: There is no evidence of sanitization or validation of the contents of the issue files before they are used to generate the QA plan or implementation steps.

super-autopilote-issue