The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several local Python scripts to manage template synchronization and health checks.
Runs python3 harness/scripts/sync-template.py with various flags (--full, --dry-run, --strategy) to perform the sync process.
Invokes python3 harness/scripts/health-check.py to diagnose project drift and update version fingerprints.
Uses python3 harness/scripts/okf-check.py to migrate legacy file formats to YAML frontmatter.
[EXTERNAL_DOWNLOADS]: Fetches structural improvements and skill definitions from an external repository.
Connects to https://github.com/Rheinmir/setup.git to sync template changes.
Uses curl and gh api to download individual files from raw.githubusercontent.com based on the manifest configuration.
[REMOTE_CODE_EXECUTION]: Downloads content from a remote source and installs it into directories where the AI agent will execute the instructions.
Installs synced files into .claude/commands/, ~/.claude/skills/, and ~/.claude/commands/.
These files serve as new or updated instructions that define the agent's future behavior.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from a remote source that is subsequently used as agent instructions.
Ingestion points: Files are downloaded from a remote GitHub repository using curl in SKILL.md (Step 4 and Step 6).
Boundary markers: The skill lacks explicit boundary markers or "ignore embedded instructions" warnings for the downloaded content.
Capability inventory: The skill can execute subprocesses via Python, write files to sensitive agent configuration directories, and perform network requests.
Sanitization: There is minimal sanitization; it only filters files that lack ## Purpose or ## Steps headers before installation.

sync-template