The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill frequently recommends and executes an external auditing tool using the command npx @andypai/harness-doctor@latest (e.g., in doctor.md, baseline.md, and docs.md). This command downloads and runs the latest version of the package from the npm registry at execution time. While the skill author's handle is abpai, the package scope @andypai is not explicitly listed as a verified vendor-owned resource in the provided context. Executing unpinned code from remote registries presents a supply chain risk.
[COMMAND_EXECUTION]: Several modules, including baseline.md and doctor.md, are designed to discover and execute shell commands from the target repository's metadata (e.g., package.json scripts, Makefiles, CI configurations). Although the skill instructs the agent to inspect command bodies and seek user confirmation, this capability allows for the execution of arbitrary code defined within the repository being analyzed.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a significant attack surface for indirect prompt injection because it ingests and processes untrusted data from the repository (README files, documentation, and manifests) to influence agent behavior and command selection. A malicious repository could contain crafted content designed to hijack the agent's logic.
Ingestion points: Modules such as baseline.md and doctor.md ingest various local files including package.json, README.md, and CI workflow files to orient the agent.
Boundary markers: The skill employs instructional boundary markers, such as advising the agent to "Inspect command bodies one hop deep" and requiring explicit user approval for destructive or external-reaching actions.
Capability inventory: The skill possesses the capability to execute shell commands, modify the filesystem, and perform network-based package installations.
Sanitization: The skill relies on human-in-the-loop confirmation and agent-based inspection rather than robust automated sanitization of ingested repository content.
[EXTERNAL_DOWNLOADS]: The skill triggers external downloads from well-known registries (npm, PyPI, etc.) via tools like npx and various package managers (uv, pip, bun, npm, pnpm) as part of its dependency hardening and auditing workflows.

harness