The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script located at ~/.claude/harness/fdk/tools/surface-coverage.py. This script is used to 'scan' the codebase to generate surface.json and 'check' coverage against coverage.json. This behavior is the primary functionality of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the user's codebase when scanning for surface items (pages, APIs, modules) and during the manual 'tracing' phase where the agent follows code paths (e.g., from a route to a controller). Malicious instructions embedded in comments or strings within the codebase could potentially influence the agent's behavior.
Ingestion points: Product code directory (<root>), domain-graph.json, and existing artifacts being checked for coverage.
Boundary markers: None explicitly implemented for reading code files, although RULE-01 mandates the use of a deterministic script for the initial surface scan.
Capability inventory: The skill can read/write files (surface.json, coverage.json), execute local shell commands via python3, and modify project files to add test gates.
Sanitization: There is no mention of sanitizing or escaping the content read from the source code before the agent processes it.

surface-coverage