The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/bootstrap_harness.py executes the git command using subprocess.run to retrieve the current commit hash. The command is invoked with static arguments and is used solely to determine if existing knowledge graphs are out of date.
[DATA_EXPOSURE]: The skill scans the target repository's files, including .env files, to identify environment variable names. The analysis logic is designed to capture only the variable keys and their locations, with explicit instructions and code implementation to avoid reading or storing the sensitive values associated with those keys.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the target repository (file contents and directory structures) to generate reports. This represents a potential surface for indirect prompt injection if the agent interprets malicious content within the repository as instructions.
Ingestion points: The bootstrap_harness.py script reads text files from the target repository provided via the --target argument.
Boundary markers: The skill creates reports in .harness/ and instructs the agent to treat these as readiness facts, but does not implement strict escaping for the content extracted from the target repo.
Capability inventory: The script performs file reads and writes within the target repository path and executes git via subprocess.
Sanitization: The script uses errors="replace" when reading files and limits scan depth, but primarily relies on the agent's internal safety filters for interpreting the generated markdown reports.

harness-bootstrap