The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves gathering and synthesizing information from untrusted external sources.
Ingestion points: The GATHER stage in SKILL.md utilizes the WebSearch and WebFetch tools to ingest content from arbitrary URLs based on the user's research request.
Boundary markers: The workflow includes a REVIEW stage (prompts defined in references/review-lenses.md) where subagents with restricted context audit the generated dossier, specifically checking if claims are supported by the archived sources.
Capability inventory: The skill can execute local scripts (archive-source.sh), write to the file system to store dossiers and source archives, and perform Git operations through automated hooks.
Sanitization: The archival script archive-source.sh attempts to mitigate risks from malicious web content by stripping script and style tags during text extraction via awk or pandoc.
[COMMAND_EXECUTION]: The skill automates workflow steps using shell scripts and Git hooks.
Evidence: The skill executes scripts/archive-source.sh to capture web content and employs several Git hooks (referenced in SKILL.md and README.md) such as sources-index-consistency.sh and dossier-commit-gate.sh. These scripts are used for repository maintenance, archival integrity checks, and ensuring that a review artifact exists before a dossier is committed.

dossier