The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled Python script scripts/import_graph.py to generate a dependency graph. It also invokes a configuration utility fitness-config.py located two levels above the skill directory, which is a common pattern for shared utilities in multi-skill environments.
[DYNAMIC_EXECUTION]: During the verification phase, the workflow instructs the agent to run project-specific code, such as CLI calls, imports, or existing tests, to confirm the impact of architectural findings. This behavior is scoped to the primary purpose of verifying architectural reachability.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code, creating an attack surface for instructions embedded in the target codebase.
Ingestion points: Source files and directories within the review target (SKILL.md Workflow Step 2).
Boundary markers: Not explicitly defined in the prompts.
Capability inventory: Writing report files to the repository (docs/architecture-review.md) and executing project-specific commands or tests (SKILL.md Workflow Step 6).
Sanitization: The skill includes an explicit instruction to avoid reproducing secret values in the report, requiring the agent to refer to them by location only.
[DATA_EXPOSURE]: The skill demonstrates security awareness by explicitly forbidding the reproduction of secret values (passwords, keys) in the generated architecture reports.

review-architecture