The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Technical.md and review-decision-register.md to identify optimization opportunities. Since these files are project-controlled, they represent an injection surface where malicious instructions could attempt to influence the agent's file modification logic during the 'Apply' phase.
Ingestion points: The skill reads Technical.md (Step 3), .candid/config.json (Steps 4 & 6), and review-decision-register.md (Step 5).
Boundary markers: None identified; the skill processes the raw text content of these files.
Capability inventory: The skill uses find, ls, git, and jq via shell execution, and utilizes Edit and Write tools to modify project files based on its analysis.
Sanitization: No explicit sanitization or validation of the text content within Technical.md or the decision register is mentioned before it is used to generate suggested edits.
[COMMAND_EXECUTION]: The skill executes several shell commands for environment discovery and validation.
Discovery: Uses find and ls to identify generated or vendored files that should be excluded (Step 4.2).
Git Integration: Executes git branch -a to detect the project's branching strategy for configuration suggestions (Step 6.3).
Validation: Uses jq to validate the integrity of JSON configuration files before attempting updates (Step 7.4).

candid-optimize