The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute the harness command-line utility, including searching for it in local directories and repo checkouts.
Evidence: The skill instructs the agent to run: harness run change-review --workspace /path/to/repo --base main --head HEAD --handoff-stdin --verbose.
Evidence: Documentation in SKILL.md mentions running roles through harness, .harness/bin/harness, or a source checkout executable.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface as it ingests and triages data from external, potentially untrusted sources produced during the review workflow.
Ingestion points: The agent is directed to read meta.json, summary.md, events.jsonl, and 'every completed reviewer JSON' as part of the triage process in SKILL.md.
Boundary markers: There are no explicit instructions or delimiters used to separate the content of these artifacts from the agent's internal instructions, nor are there warnings to ignore embedded directives.
Capability inventory: The skill grants the agent the ability to execute shell commands (harness) and modify the local codebase based on the triaged information.
Sanitization: The process does not specify any validation or sanitization of the artifact content before it is used by the agent to determine implementation dispositions (e.g., 'Implement', 'Adapt').

change-review-workflow