The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes implementation plans and subagent reports, creating a surface where malicious instructions in these inputs could influence subagent behavior.
Ingestion points: SKILL.md describes reading plans (e.g., docs/plans/feature-plan.md) and status reports from subagents (implementers and reviewers).
Boundary markers: Instructions emphasize creating a "fresh subagent per task" with "isolated context" and explicit instructions to redact secrets/PII before sharing context with subagents.
Capability inventory: Tools for file system access, git operations, and spawning new subagents.
Sanitization: The framework includes a mandate to redact sensitive data before sharing context between agents.
[DYNAMIC_EXECUTION]: The skill dynamically generates prompts for subagents using provided templates.
Pattern: Simple prompt generation from known templates using task-specific data.
Evidence: Templates in implementer-prompt.md, spec-reviewer-prompt.md, and code-quality-reviewer-prompt.md are filled with content at runtime to orchestrate subagent tasks.

subagent-driven-development