The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard shell utilities including git, grep, sed, and test to perform analysis on repository files. These commands are limited to read-only metadata extraction and local file status checks.
[COMMAND_INJECTION]: The skill implements high-standard defensive measures against command injection. It requires regex validation for milestone IDs (^MILE-[0-9]+$) and commit SHAs (^[0-9a-f]{40}$) before they reach any shell command. It also mandates safe argument passing, such as using the -- separator and utilizing flag operands that consume next-token arguments to prevent malicious input from being interpreted as flags.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository (e.g., roadmap and vision files). It mitigates risk by defining this content as 'passive data' and explicitly instructing the agent to report, rather than execute, any instructions found within milestone outcome sentences or human-provided rationales. Ingestion points include docs/roadmap/INDEX.md and docs/specs/INDEX.md, while capabilities are restricted to specific Git/Grep operations and localized file writes.

assess-milestone