The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs local environment audits and file management using the Bash tool.
Evidence: Step 1 executes shell loops to iterate through the .claude/skills/ directory, extracting version strings with grep and checking file status with git status.
Evidence: Step 5a executes commands to create a backup directory ops/skills-archive and copy existing skill files to it.
Evidence: Step 6 uses bash and grep to verify the structural integrity of the updated skill files.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from various project and plugin files, which serves as a surface for potential indirect instructions.
Ingestion points: Reads project configuration (ops/derivation.md, ops/config.yaml), existing skill definitions (.claude/skills/*/SKILL.md), and research documents from the plugin root (${CLAUDE_PLUGIN_ROOT}/methodology/).
Boundary markers: The skill does not implement specific delimiters or "ignore embedded instructions" warnings when parsing these external files.
Capability inventory: The skill possesses Read, Write, Edit, and Bash tools, which are necessary for its upgrade function but could be targets for injection.
Sanitization: No explicit sanitization or filtering logic is provided for the content read from the files before it is used to generate proposed updates.
This surface is considered a low risk due to the mandatory human-in-the-loop review process and the primary purpose of the skill as a maintenance utility.

upgrade