The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from project documentation files located in the docs/ directory.
Ingestion points: Reads candidate documentation from docs/acceptances/ATTENTION.*.md and docs/<feature>/ACCEPTANCE.md.
Boundary markers: No explicit delimiters are used, but the instructions emphasize user control over the content.
Capability inventory: Modifies documentation files, deletes local feature files, and invokes a local commit skill (../commit-changes/SKILL.md).
Sanitization: Implements a mandatory human-in-the-loop workflow where the user must approve or provide the exact wording for every entry before it is promoted to the main file. This prevents the agent from autonomously acting on or being influenced by instructions potentially embedded in the documentation.
[SAFE]: The skill's operations are limited to project-specific documentation and follow a standard, user-vetted workflow. It does not perform network operations, access sensitive system files, or utilize obfuscation techniques.

end-acceptance