The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements an indirect prompt injection surface by consuming and acting upon the FIX_MANIFEST block within an external audit-report.md file. Maliciously crafted manifest items could influence the agent to perform unintended file operations or embed deceptive instructions into project documentation.
Ingestion points: .claude/audit-report.md (specifically the FIX_MANIFEST section).
Boundary markers: The skill looks for a specific YAML-like block within the markdown file.
Capability inventory: The skill has the ability to create directories (openspec/), create and modify project context files (CLAUDE.md, ai-context/*.md), and recursively delete local skill directories (.claude/skills/).
Sanitization: No explicit sanitization or validation of the manifest content is described beyond basic normalization and a compatibility policy for action types.
[COMMAND_EXECUTION]: The skill performs automated file system operations including directory creation and the recursive deletion of skill subdirectories (delete_duplicate handler). Although these operations require user confirmation at each phase, they grant the skill significant control over the local .claude/ environment based on external input.

project-fix