The Agent Skills Directory

[COMMAND_EXECUTION]: The README.md instructions recommend executing PowerShell with the -ExecutionPolicy Bypass flag, which is a privilege escalation pattern that disables standard script execution restrictions on Windows systems.
[COMMAND_EXECUTION]: The scripts/install-coherence-hooks.sh script establishes persistence within the development environment by installing Git hooks (pre-commit or pre-push) that automatically trigger the coherence checking logic during standard repository operations.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it is designed to ingest and reconcile untrusted external prompt data through the coherence-plan.mjs script via the --prompt-file argument.
Ingestion points: scripts/coherence-plan.mjs reads the content of a user-specified file and scripts/check-coherence.mjs analyzes the local repository state.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat the external file content as untrusted data during processing.
Capability inventory: The skill possesses the ability to write multiple markdown strategy files, modify Git hooks, and change file permissions (chmod).
Sanitization: While filename generation is sanitized using a slug generator, the ingested prompt content is interpolated directly into generated documentation files.
[DATA_EXFILTRATION]: The scripts/coherence-plan.mjs script facilitates data exposure by reading the contents of arbitrary local files provided through the --prompt-file argument, which could be exploited to read sensitive configuration or credential files.
[EXTERNAL_DOWNLOADS]: The README.md documentation provides instructions for installing the skill directly from the vendor's GitHub repository using npx or yarn dlx, which downloads external code for execution.

coherence