feat: /implement — delegate reviewed plan → code to OpenAI Codex#1833
Closed
merit-blake wants to merge 2 commits into
Closed
feat: /implement — delegate reviewed plan → code to OpenAI Codex#1833merit-blake wants to merge 2 commits into
merit-blake wants to merge 2 commits into
Conversation
Adds /implement, a Claude-host-only skill that takes a reviewed plan (from /plan-eng-review or /autoplan) and hands implementation to OpenAI Codex via `codex exec` running in the background on a cc worktree. Polls codex's JSONL stream for progress, then auto-runs /codex review + /review as an independent gate before handing back. P1 findings boomerang back to codex for a fix pass (capped at 3). Operating model: codex runs with -s danger-full-access. The cc worktree is the safety boundary (branch-isolated, throwaway via git worktree remove); the filesystem-boundary instruction in the prompt does the meaningful work (stops codex from reading agent skill files under ~/.claude, ~/.codex, etc). State (logs, prompts, summary, decisions) lives at ~/.gstack/projects/$SLUG/implement/$BRANCH/ — branch-scoped so concurrent runs on sibling worktrees don't collide, survives the worktree being deleted, gets gbrain-sync'd. Plan discovery walks gstack project dir, ceo-plans, ~/.claude/plans, ~/.codex/plans, .gstack/plans, plans/. Plan clearance uses gstack-review-read (the mechanism /ship uses). Skipped from non-Claude hosts (codex, factory, gbrain, hermes, openclaw, kiro, opencode, cursor, slate) — same pattern as /codex, since this is a Claude wrapper around codex exec. Hand-maintained registries (README, docs/skills.md, AGENTS.md) and test exclusion lists updated to match. Re-homed onto current main from a stale reference branch (~39 commits behind); only the skill source and host-gating intent were ported. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Output of `bun run gen:skill-docs` after adding implement/SKILL.md.tmpl: the generated implement/SKILL.md, the /implement entry in gstack/llms.txt, and the routing block in scripts/proactive-suggestions.json. No other generated files drifted (root SKILL.md and AGENTS.md registry untouched). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Author
|
Closing — opened in error, reverting. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this adds
Re-homes the
/implementskill onto currentmain./implementis aClaude-host-only skill that takes a reviewed plan (from
/plan-eng-reviewor/autoplan) and hands implementation to OpenAI Codex viacodex execrunningin the background on a cc worktree. It polls Codex's JSONL progress stream, then
auto-runs
/codex review+/reviewas an independent gate before handing back.P1 findings boomerang back to Codex for a fix pass, capped at 3.
-s danger-full-access; the cc worktreeis the safety boundary (branch-isolated, throwaway). The filesystem-boundary
instruction in the prompt keeps Codex from reading agent skill files.
~/.gstack/projects/$SLUG/implement/$BRANCH/— branch-scoped, survives theworktree being deleted, gbrain-sync'd.
hermes, openclaw, kiro, opencode, cursor, slate) via
skipSkills, samepattern as
/codex.How it was built
Re-homed onto current
mainfrom a stale reference worktree (~39 commitsbehind). Only the skill source (
implement/SKILL.md.tmpl) and host-gatingintent were ported; the stale branch was not merged. Generated docs were
produced by the current
bun run gen:skill-docspipeline, not copied from thestale branch.
Commits are bisected:
feat:— skill template, host-gating, test exclusion lists, hand-maintainedregistries (README / docs/skills.md / AGENTS.md).
chore:— regeneratedimplement/SKILL.md,gstack/llms.txt,scripts/proactive-suggestions.json.Tests
bun test→ exits 0. Targeted:skill-validation+gen-skill-docs=719 pass, 0 fail.
Cannot find package 'sharp'warning surfaces inbrowse/test/screenshot-size-guard.test.ts(missing dev dependency in thisworktree); unrelated to these changes and does not fail the suite.
Notes
/ship(per therepo's "write the CHANGELOG at ship time" rule).
merit-blake/gstack) because the author lacks writeon the base repo. Fork PRs don't receive base-repo secrets, so the eval/E2E
CI jobs will fail with empty-env auth; check-freshness, workflow-lint, and
windows-tests should still pass. To get eval CI, a maintainer can push the
branch to the base repo and re-target.
🤖 Generated with Claude Code