The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several shell commands and local scripts to perform maintenance tasks.
Evidence includes the use of git submodule update, git diff, python3 scripts/check-provenance.py, pnpm exec vp run site:build, and scripts/check-all.sh.
These commands are functional components of the skill's purpose to manage code drift and build project documentation.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from git submodules and instructions suggest porting changes directly into the local environment, which could ingest adversarial content.
Ingestion points: Git submodules tracked in source-skills/, as well as repository files such as CHANGELOG.md and GLOSSARY.md.
Boundary markers: No specific prompt boundaries or delimiters are mentioned to prevent the agent from obeying instructions that might be contained within the diffs being processed.
Capability inventory: The skill has access to shell execution, file system diffing, and local script execution (check-all.sh, check-provenance.py).
Sanitization: The skill mitigates risk by instructing the agent to leave changes uncommitted for human review and to prompt the user if the intent of a change is unclear.

update-from-source