The Agent Skills Directory

[COMMAND_EXECUTION]: The skill extensively uses shell commands to manage git worktrees and project lifecycles. It utilizes pluggable execution hooks via environment variables such as ${VALIDATION_TEST_CMD}, ${VALIDATION_DEPLOY_CMD}, and ${POST_MERGE_CLEANUP_CMD}. These allow the agent to execute project-specific test and build suites, which is standard behavior for an engineering orchestrator.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes deliverables and reports from various sub-agents.
Ingestion points: Untrusted data enters the context through sub-agent handoff messages, Git branch status, and Orca delivery logs (documented in SKILL.md).
Boundary markers: The skill implements a structured '4-Part Handoff Summary' and requires explicit 'Scope & Intent' briefings to delimit sub-agent input. Manual review gates are enforced by default for UI/UX changes.
Capability inventory: The skill can perform file system operations (git worktrees), shell command execution (via npm and orca), and local network requests (probes to localhost:3111).
Sanitization: The Coordinator is instructed to inspect logs, screenshots, and evidence before merging changes, providing a validation layer against malicious input.
[DATA_EXPOSURE]: The scripts/detect-env.sh script probes for system binaries and manages a local configuration file in the user's home directory. This activity is transparent, standard for tool detection, and does not involve exfiltration to external domains.

multi-agent-orchestration