The Agent Skills Directory

[SAFE]: The skill implements significant safety guardrails for code modification tasks. It requires a baseline check before any fix, mandates --dry-run for previews, and requires manual review of git diff after execution. It explicitly prohibits modifying business semantics, security thresholds, or dependencies without specific authorization.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it processes untrusted data from linter logs and source code files during the fix process.
Ingestion points: Reads outputs from local tools like bin/codeguard check, eslint, and npm audit, as well as file content through git diff.
Boundary markers: The skill instructions emphasize 'Saving Baseline' (Step 1) and 'Reviewing Diff' (Step 5) as critical boundaries, though specific prompt-level delimiters for tool outputs are not explicitly defined in the provided markdown.
Capability inventory: The skill is capable of executing shell commands (bin/codeguard fix, npm audit fix, eslint --fix) and reading/writing to the local filesystem.
Sanitization: The workflow relies on manual human review of differences (git diff review) and re-running checks to ensure no unintended changes occurred, providing a validation layer against malicious input manipulation.

codeguard-fix