The Agent Skills Directory

[SAFE]: The skill and its associated reference documents (Phase 1 through 8) are purely instructional. They define a robust software engineering process for building features in a layered architecture. Analysis detected no obfuscation, malicious commands, hidden payloads, or attempts to exfiltrate data.
[PROMPT_INJECTION]: The 'Discovery' phase (references/01-discovery.md) instructs the agent to read existing project documentation (e.g., docs/features/*.md) end-to-end and treat it as a contract. This presents a potential surface for indirect prompt injection if an attacker can commit malicious instructions to the documentation directory.
Ingestion points: Project-level documentation files in docs/architecture/, docs/features/, and CLAUDE.md.
Boundary markers: Absent; the agent is instructed to treat the documents as authoritative for the design.
Capability inventory: Writing Swift source code (Phase 3), updating documentation (Phase 7), and executing shell commands for testing and building (Phase 8).
Sanitization: Absent. While this surface exists, it is a functional requirement for the skill's intended purpose and is handled with low risk given standard agent guardrails.
[COMMAND_EXECUTION]: Phase 8 (references/08-verify.md) directs the agent to execute build and test commands (e.g., swift test, xcodebuild). These are standard tools for Swift development and are used in a legitimate context to verify the implemented feature.

feature-lifecycle