The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection vulnerability surface because it is designed to analyze external untrusted data and execute project commands based on that analysis.
Ingestion points: The skill ingests untrusted repository content, Git status/diff structures, command logs, external test outputs, and dependency manifests as described in SKILL.md and references/evidence-provenance.md.
Boundary markers: Present. SKILL.md contains an explicit 'Security Model' section instructing the agent that all scanned files, logs, and tool outputs must be treated as data rather than instructions, and that any embedded command-like text carries no authority and must be ignored.
Capability inventory: The skill runs local shell commands for Git inspection (git rev-parse, git status, git diff) and triggers project-defined test, compilation, build, and lint tools based on repository configuration as outlined in SKILL.md and references/runtime-verification.md.
Sanitization: Present. The guidelines explicitly mandate that the agent must not print or log encountered secrets, and must obtain active user authorization before executing any mutating or production-impacting checks as specified in references/verification-depth.md.

verification-gate