The Agent Skills Directory

[SAFE]: The skill's primary function is conversational requirement gathering and artifact generation (markdown files). It explicitly states that the interviewer does not implement features, which significantly reduces the risk of unauthorized code changes.
[SAFE]: The workflow incorporates a 'no-write gate' during the planning phase and requires a single, positive human checkpoint before any repository persistence occurs, ensuring that all filesystem operations are user-verified.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection due to its ingestion of external repository data.
Ingestion points: SKILL.md (Workflow step 2) and references/workflow-details.md (Interview step 2) instruct the agent to inspect repository context including issue descriptions, README files, and existing ADRs.
Boundary markers: Present. The assets/codex-execution-prompt.md template uses <implementation_spec> tags to delimit potentially untrusted content.
Capability inventory: The skill is capable of repository file writes (artifact-destinations.md), but these are strictly limited to documentation paths and require manual user approval.
Sanitization: Absent, although the 'Single Checkpoint' workflow ensures human review of all generated content before it is processed or saved.
[SAFE]: All external documentation references provided in the skill (e.g., in references/host-adapters.md) point to well-known and trusted developer domains such as OpenAI, Anthropic, and Cursor.

codex-spec-interviewer