The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates a multi-agent workflow where a parent agent processes evidence from subagents that ingest potentially untrusted repository content, introducing a surface for indirect prompt injection.\n
Ingestion points: Subagents are instructed to scan the repository using Read, Grep, and Glob tools as defined in the role stubs in references/agent-templates.md.\n
Boundary markers: The skill mandates that subagents start with no parent transcript history ("subagents inherit NO parent transcript") to limit context exposure, as specified in SKILL.md.\n
Capability inventory: Depending on the role, agents have access to Bash, Edit, and Write tools.\n
Sanitization: The instructions require the parent agent to manually synthesize returns and resolve conflicts between agents, providing a human-in-the-loop oversight mechanism.\n- [COMMAND_EXECUTION]: The provided templates grant subagents access to the Bash tool for repository searching and verification tasks.\n
Evidence: The role definitions for repo-explorer, reviewer, and implementation-worker in references/agent-templates.md explicitly include the Bash tool in their tool sets to support inventory scans and test execution.

fan-out