The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external Git repositories, which may contain malicious instructions hidden in documentation or source code.
Ingestion points: Reads file content across multiple user-provided repository paths, including AGENTS.md, CLAUDE.md, and README files.
Boundary markers: The skill uses git status snapshots to establish preservation boundaries but does not specify delimiters for isolating read content from instructions.
Capability inventory: The skill can execute shell commands, run repository-defined build scripts, and perform file writes/commits.
Sanitization: There is no explicit content sanitization, but the skill requires mechanical verification of all findings and mandatory user consent before implementation.
[DYNAMIC_EXECUTION]: The skill is instructed to execute build pipelines, regeneration commands, and Git hooks defined within the audited repositories.
It identifies and runs regeneration artifacts through build-only paths to verify drift.
Executing scripts found within the repositories constitutes dynamic execution of content provided by the user (or a third party if the repo is not the user's own).
[COMMAND_EXECUTION]: The skill heavily utilizes shell commands for repository management and investigation.
Uses git status --short, git diff, and path resolution commands.
Supports tilde expansion for repository paths provided in $ARGUMENTS.

repo-harmonization