The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process external requirements data from GitHub issues and user tasks to generate specifications and implementation tickets.
Ingestion points: Untrusted data enters the agent context through user-supplied tasks and existing GitHub issue bodies/comments read via the github-facts.mjs script.
Boundary markers: The skill uses specific Markdown headers (e.g., ## Affected seams, ## Acceptance criteria) and stable local IDs (e.g., AC-1) to identify authoritative content, which are validated by structural logic in workflow-state.ts.
Capability inventory: The skill has the ability to read local files, execute bundled scripts, and perform authenticated read/write operations on GitHub via the gh CLI.
Sanitization: Ingested content is parsed and validated against expected templates using the validateAuthoritativeBody and resolveRequirementsBody functions to prevent malformed data from being treated as authoritative requirements.
[COMMAND_EXECUTION]: The skill uses bundled Node.js scripts (github-facts.mjs and workflow-cli.mjs) to perform GitHub operations and state validation. These scripts use execFileSync to invoke the gh (GitHub CLI) tool. Although the script validates parameters such as repository names and issue numbers using regular expressions, the execution of shell commands based on data derived from the agent's interaction with the environment is a noteworthy security surface.
[EXTERNAL_DOWNLOADS]: The INSTALL.md file provides instructions for installing the skill and its dependencies from external GitHub repositories, including the mattpocock/skills collection. These references are documented neutrally as they occur during the user-initiated installation process and are not executed dynamically at runtime.

plan-this