The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local git commands and a provided shell script to detect the development environment and apply configuration settings.
Evidence: Runs git remote -v, git config --local, and scripts/discover-toolchain.sh to probe for tools and repository metadata.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes local repository files to configure the agent environment, creating a potential surface for indirect injection if these files are maliciously crafted.
Ingestion points: scripts/discover-toolchain.sh parses package.json, Cargo.toml, go.mod, pubspec.yaml, and other manifests during the Phase 1 discovery process.
Boundary markers: The skill implements a Phase 2 "Staging & Authorization Gate" where all proposed changes are presented as a Changeset for human approval before application.
Capability inventory: The skill can write files to the .agents/ and docs/ directories, modify .gitignore, and update local .git/config settings.
Sanitization: The discovery script uses specific pattern matching for manifest files, and the mandatory human review of the Changeset serves as the primary defense against malicious data poisoning.

zenf-init