The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/kgr_github.py executes the GitHub CLI (gh) via subprocess.run() to perform GraphQL and REST API queries. The implementation uses a list of arguments and relies on the default shell=False behavior, which is a safe practice to prevent shell command injection.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from the GitHub API, such as pull request status, check run conclusions, and release asset names. While this data originates from external sources, the skill implements several mitigations to prevent indirect prompt injection:
Ingestion points: Data enters the system via the gh CLI in the pr_snapshot, workflow_snapshot, and tag_snapshot functions in scripts/delivery_wait.py.
Boundary markers: External data is parsed and normalized into structured JSON observations, maintaining clear boundaries between data and logic.
Capability inventory: The skill has the capability to execute shell commands (via the gh wrapper) and write files to the local file system (for checkpointing state).
Sanitization: As documented in SKILL.md and implemented in the snapshot functions, checkpoints explicitly exclude credentials, raw comment bodies, and raw API payloads, ensuring that potentially malicious content from external sources is not persisted or processed as instructions.

delivery-wait