forge-loop

Prompt file imported from FlexNetOS/flexnetos_runner (.codex/prompts/forge-loop.md). Fill in {{arguments}} before use. Copyright stays with the author.

Use the repository's Rust forge-loop engine instead of improvising the loop in chat.

Command:

fxrun forge-loop run --goal "{{arguments}}"

Rules:

• Follow TDD: prove or create a red test before implementation.
• Run self-evaluation on every cycle.
• Mine the required target resources before claiming an upgrade: OpenAI Codex GitHub Action, Permissions, Subagents, RoggeOhta/awesome-codex-cli, and Yeachan-Heo/oh-my-codex.
• Keep auto-compaction enabled for every local, action, and nested codex exec session; preserve the active phase, source matrix, validation state, and next action in compact summaries.
• Record source-attributed findings and map each applied change to a config, hook, rule, skill, subagent, permission, GitHub Action, or tool surface.
• Use isolated named worktrees for concurrent loop work; follow .codex/worktrees/forge-loop-isolation.toml and do not run multiple mutating loops in the same checkout.
• Spawn project-scoped subagents only for genuinely parallel research/review/CI audit work; wait for all results and synthesize evidence before editing.
• Use research findings to improve reliability, accuracy, and speed.
• Commit, push, open a PR, and auto-merge green PRs when repository settings allow.
• Strict upgrade only: no downgrade/removal unless a replacement is installed, configured, and parity-proven. Complete .codex/checklists/forge-loop-cycle.toml evidence before claiming a cycle done.
• Before completion, run fxrun forge-loop components-audit --strict, fxrun forge-loop target-mining-audit --strict, and fxrun forge-loop docs-drift --json; if Codex config surfaces changed, update the component audit and CI guard in the same PR.