The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests and processes plan artifacts (e.g., plan.md and task files) that could contain malicious instructions designed to influence subagent behavior or the orchestration logic.\n
Ingestion points: The skill reads plan artifacts from the docs/threads/ lineage folder and passes task-specific content to the implementer and reviewer subagents in their respective briefs (SKILL.md, Workflow Step 6).\n
Boundary markers: The orchestrator uses isolated subagent briefs and scratch workspaces (.wip/) to delimit context, although it does not specify explicit platform-level delimiters for data-vs-instruction separation.\n
Capability inventory: The skill performs git add and git commit operations (SKILL.md) and enables subagents to modify the project's source code. It also executes verification commands defined within the plan artifacts (references/plan-compliance-reviewer.md).\n
Sanitization: This risk is significantly mitigated by a mandatory merged reviewer subagent that performs independent plan-compliance and code-quality checks before any changes are committed.\n- [COMMAND_EXECUTION]: The skill utilizes shell commands to manage the repository state and automate the verification process.\n
Evidence: The skill executes standard git operations including git status, git diff, git add, and git commit to maintain the audit trail (SKILL.md). It also executes project-specific scripts and verification blocks (e.g., npm test, grep, test -f) as prescribed by the plan and project requirements (references/plan-compliance-reviewer.md, references/code-quality-reviewer.md).

implement-plan-with-subagents