The Agent Skills Directory

[COMMAND_EXECUTION]: The utility scripts (e.g., change-archive.mjs, context-verify.mjs) use child_process.execSync to interact with the system's git and zip binaries. These operations are used for legitimate development tasks such as identifying code drift and zipping completed change artifacts. The input to these commands, specifically the change identifiers, is sanitized using a regex that restricts characters to alphanumeric, hyphens, and spaces, preventing shell injection attacks.
[INDIRECT_PROMPT_INJECTION]: As a tool designed to analyze and modify source code and architectural documentation (CONTEXT.md), the skill inherently processes untrusted data.
Ingestion points: SKILL.md instructions and context-discover.mjs read project files and CONTEXT.md documentation into the agent's context.
Boundary markers: The provided templates and protocols (e.g., plan.md.tmpl) enforce the use of functional specifications rather than executable code snippets, acting as a structural delimiter.
Capability inventory: The skill can write and delete files within the .changes/ directory and execute local commands (git, zip) for workflow management.
Sanitization: The lib/index.mjs utility provides centralized sanitization for change IDs used in filesystem paths and command arguments, mitigating the risk of path traversal or injection from malicious project titles.

triage