The Agent Skills Directory

[SAFE]: The skill implements a professional research framework focused on cited evidence and verifiable traces. It includes comprehensive Bash scripts for validating the integrity and provenance of its output artifacts, ensuring they are bound to specific Git commits.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it is designed to ingest and analyze untrusted repository content and external reports.
Ingestion points: Source code files, build manifests, external facts, and caller-supplied reports are processed as part of research activities (SKILL.md, references/research.feature).
Boundary markers: The instructions mandate the separation of observations, inferences, and contradictions. They specifically instruct the agent to mark unverified external claims as 'unknown' and to verify search hits against actual sources.
Capability inventory: The skill utilizes local shell scripts (scripts/codebase-recon/validate-output.sh) that execute standard system tools including git, jq, grep, stat, and sha256sum. It also allows for the use of host-provided web tools for verifying external facts.
Sanitization: Validation scripts perform integrity checks using SHA-256 digests and Git history verification. While no content-level sanitization for LLM instructions is present, the skill relies on strict evidentiary rules to mitigate the influence of malicious data.

research