
from recursive-mode52
Orchestrates recursive repository workflows: phased runs, locks, audits, and durable memory updates for long-running agentic tasks.
This skill provides a full workflow orchestration layer for repositories that adopt a recursive-mode convention. It defines canonical run roots, phase gates, lock verification, and memory/decision ledgers so agents can execute multi-phase work (draft → audit → repair → pass → lock) and record durable artifacts.
Use when you need to run or resume a staged implementation in a repo that follows the recursive-mode scaffold (e.g., running phases, verifying locks, generating review bundles, or updating skill memory). It's intended for agent-driven project runs, audits, and delegated subagent orchestration.
Best used by agents with repository and CLI access (Codex/Copilot-style or CLI-capable agents). Works well with subagent routing systems and agents that can run Python/Bash scripts.
Recursive-mode is a heavyweight workflow orchestration skill for phased AI agent implementation with 72 scripts (Python + PowerShell). It manages run directories, phase locking, audits, and subagent routing. Only 5 of 37 non-skipped scripts ran successfully — most failed due to ModuleNotFoundError when scripts couldn't find sibling modules. The SKILL.md is extremely detailed but verbose. No security issues found; the --dangerously-bypass flag is a Codex CLI test parameter, not a production bypass.
recursive_router_lib (cross-script import resolution requires sys.path setup)No security concerns. The 'dangerously-bypass-approvals-and-sandbox' flag is used only in the benchmark script for testing OpenAI Codex CLI, not in production code. PowerShell 'Bypass' is ExecutionPolicy Bypass for tests, not a security bypass. Scripts use subprocess with list args (safe). Well-engineered but over-engineered for a skill — very niche audience.