The Agent Skills Directory

[PROMPT_INJECTION]: The skill's workflow involves reading implementation plans and task descriptions and interpolating them into specialized subagent prompts (Ingestion points: SKILL.md, implementer-prompt.md, spec-reviewer-prompt.md). It uses markdown headers as delimiters (Boundary markers: present) and provides subagents with extensive capabilities to modify code, execute tests, and commit changes (Capability inventory: file system access, test execution, git operations). No formal sanitization or validation of the input plan text is performed (Sanitization: absent). This configuration creates an inherent surface for indirect prompt injection, where a malicious implementation plan could attempt to override subagent instructions.
[COMMAND_EXECUTION]: The skill dispatches subagents to perform development tasks such as writing code and running tests. These operations are conducted within the local workspace using internal platform tools and are appropriate for the skill's stated purpose.
[DATA_EXFILTRATION]: No evidence of unauthorized network activity or attempts to exfiltrate sensitive information (such as API keys or SSH credentials) was found.
[EXTERNAL_DOWNLOADS]: The skill does not download external scripts or execute code from untrusted remote sources during its operation.

subagent-driven-development