The Agent Skills Directory

[PROMPT_INJECTION]: No malicious patterns or bypass attempts were detected. The workflow instructions are strictly procedural and do not attempt to override agent safety guidelines.
[DATA_EXFILTRATION]: No unauthorized data access or exfiltration patterns were found. The skill operates on local development files and does not perform network operations to external or untrusted domains.
[INDIRECT_PROMPT_INJECTION]: The skill processes implementation plans which are potentially untrusted data. This architectural risk is handled by the multi-stage review process.
Ingestion points: Implementation plan files (e.g., docs/plans/feature-plan.md) read during the task extraction phase as described in SKILL.md.
Boundary markers: Prompt templates in implementer-prompt.md utilize markdown headers to separate the plan content from the subagent's instructions.
Capability inventory: Dispatched subagents utilize general-purpose and code-reviewer tools which allow for code creation and verification.
Sanitization: No explicit automated sanitization is performed; the workflow relies on the 'spec compliance' and 'code quality' reviews to ensure the implementation is safe and accurate.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns were detected. The skill relies on standard local tools for code development tasks and does not download or execute scripts from external URLs.

subagent-driven-development