From cloclo
CLoClo — Claude + Codex Collaboration pipeline. Orchestrates the full dev cycle by chaining SuperPowers skills (brainstorming, writing-plans, subagent-driven-development, verification-before-completion) with Codex reviews between phases. Triggers on: /pipeline, new feature, implement, build, create
How this skill is triggered — by the user, by Claude, or both
Slash command
/cloclo:pipelineThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Orchestrate the full development cycle by combining **existing skills** with
Orchestrate the full development cycle by combining existing skills with Codex review phases and interactive decision points.
CLoClo does NOT reimplement brainstorming, planning, execution, or verification. It invokes the real skills and adds Codex reviews + decision points between them.
At pipeline start, check and automatically fix missing dependencies.
# Try to invoke a SuperPowers skill to see if it's available
If SuperPowers is NOT available:
~/.claude/settings.json"superpowers@superpowers-marketplace": true to enabledPlugins"superpowers-marketplace": {"source": {"source": "github", "repo": "obra/superpowers-marketplace"}} to extraKnownMarketplaces (if not already present)settings.json/pipeline again."codex --version
If codex is NOT found:
npm install -g @openai/codexnpm install -g @openai/codex"codex --version to confirmcodex whoami
If not authenticated:
! codex login (the ! prefix runs it in this session)."find ~/.claude/plugins -name codex-companion.mjs -path '*/codex/scripts/*' 2>/dev/null | head -1
If companion NOT found:
~/.claude/settings.json"codex@openai-codex": true to enabledPlugins"openai-codex": {"source": {"source": "github", "repo": "openai/codex-plugin-cc"}} to extraKnownMarketplaces (if not already present)settings.json/pipeline again."If Codex CLI, auth, or plugin fail but SuperPowers is available:
docs/cloclo-sessions/YYYY-MM-DD-<slug>/
<slug> = 2-3 word kebab-case summary of the topicsession.log:
[timestamp] CLoClo session started: <slug>
[timestamp] Prerequisites: superpowers=OK, codex=OK|MISSING, auth=OK|MISSING
pipeline.config.md
in the session directory (ask the user or auto-detect from project files).superpowers:brainstormingInvoke Skill("superpowers:brainstorming")
This gives you the FULL SuperPowers brainstorming experience:
After brainstorming completes: The spec will be in docs/superpowers/specs/YYYY-MM-DD-*-design.md.
Copy or symlink it to {session_dir}/01-spec.md for session tracking.
Log: [timestamp] Phase 1 complete: 01-spec.md
GATE: Do not proceed until the user has explicitly approved the spec.
If Codex available: Invoke the codex-review skill with:
review_type: specinput_file: {session_dir}/01-spec.mdoutput_file: {session_dir}/02-codex-review-spec.mdIf Codex unavailable: Skip with warning, go directly to Decision Point #1.
Log: [timestamp] Codex review spec: {status}
Present the Codex findings (raw, no filtering) and ask:
Codex a review ta spec. Voici ses findings :
[3-5 line summary of key findings from 02-codex-review-spec.md]
Fichier complet : {session_dir}/02-codex-review-spec.md
Que veux-tu faire ?
A. Integrer tous les findings et continuer
→ SuperPowers corrige la spec, ecrit 03-spec-v2.md
B. Integrer certains findings (tu precises lesquels)
C. Ignorer la review et continuer avec la spec actuelle
D. Demander a Codex de creuser un point precis
→ Codex repart en review ciblee
E. Modifier la spec toi-meme
→ Tu edites le fichier, dis "c'est bon" quand pret
Ou tape ce que tu veux — commentaire libre.
Log the user's choice in session.log.
After A or B: SuperPowers corrects the spec → {session_dir}/03-spec-v2.md.
Do NOT automatically re-submit to Codex. User controls re-review via D.
superpowers:writing-plansInvoke Skill("superpowers:writing-plans")
Input: the approved spec (01-spec.md or 03-spec-v2.md).
This gives you the FULL SuperPowers plan writing:
After plan is written: The plan will be in docs/superpowers/plans/YYYY-MM-DD-*.md.
Copy or symlink it to {session_dir}/04-plan.md.
Log: [timestamp] Phase 3 complete: 04-plan.md
If Codex available: Invoke codex-review with:
review_type: planinput_file: {session_dir}/04-plan.mdoutput_file: {session_dir}/05-codex-review-plan.mdspec_path: path to approved specIf Codex unavailable: Skip with warning.
Same A-E format as Decision Point #1.
If corrections: SuperPowers rewrites → {session_dir}/06-plan-v2.md.
superpowers:subagent-driven-developmentInvoke Skill("superpowers:subagent-driven-development")
Input: the approved plan (04-plan.md or 06-plan-v2.md).
This gives you the FULL SuperPowers execution:
Log: [timestamp] Phase 5 complete: [commit list]
Record base_ref (git SHA before execution) and commit_list (all new commits).
If Codex available: Invoke codex-review with:
review_type: implinput_file: {session_dir}/04-plan.md (or 06-plan-v2.md)output_file: {session_dir}/07-codex-review-impl.mdspec_path: approved specplan_path: approved planbase_ref: git SHA recorded before Phase 5commit_list: all commits from Phase 5If Codex unavailable: Skip with warning.
Codex a review l'implementation. Voici ses findings :
[Summary of findings]
A. Integrer tous les findings → SuperPowers corrige, nouveau commit
B. Corriger certains findings (tu precises)
C. Ignorer et passer a la verification
D. Demander a Codex de creuser un bug potentiel
E. Lancer un audit complet du projet
Ou commentaire libre.
superpowers:verification-before-completionInvoke Skill("superpowers:verification-before-completion")
This gives you the FULL SuperPowers verification:
If pipeline.config.md exists in the session, use its verification commands.
Otherwise, let the verification skill auto-detect or ask the user.
Log: [timestamp] Phase 7 complete: {PASSED|FAILED}
This phase runs only if the implementation touched UI files (.tsx, .jsx, .vue, .svelte, .html, .css).
If no UI files were modified, skip to Phase 8.
command -v agent-browser >/dev/null 2>&1
If agent-browser is NOT found:
[timestamp] Phase 7.5 SKIPPED: agent-browser not availableIf agent-browser IS available:
Detect if any commits from Phase 5 modified UI files:
git diff --name-only {base_ref}..HEAD | grep -E '\.(tsx|jsx|vue|svelte|html|css|scss)$'
If UI files found, identify the affected pages/routes.
For each affected page:
agent-browser open <url> # Open the page
agent-browser snapshot # Get interactive elements
agent-browser screenshot <path> # Capture visual state
Read and verify EVERY screenshot immediately. Do not skip this.
Save verification screenshots to {session_dir}/screenshots/ for evidence.
If issues found:
Rules:
Log: [timestamp] Phase 7.5 complete: {N pages verified, M screenshots}
docs/cloclo-sessions/YYYY-MM-DD-<slug>/
├── 01-spec.md ← From superpowers:brainstorming
├── 02-codex-review-spec.md ← From codex-review skill
├── 03-spec-v2.md ← If corrections after review
├── 04-plan.md ← From superpowers:writing-plans
├── 05-codex-review-plan.md ← From codex-review skill
├── 06-plan-v2.md ← If corrections after review
├── 07-codex-review-impl.md ← From codex-review skill
├── screenshots/ ← Visual verification evidence (Phase 7.5)
│ ├── page-name-01.png
│ └── page-name-02.png
├── session.log ← All decisions + timestamps + job-ids
└── pipeline.config.md ← Optional verification config
[2026-04-06T14:30:00] CLoClo session started: auth-refactor
[2026-04-06T14:30:01] Prerequisites: superpowers=OK, codex=OK, auth=OK
[2026-04-06T14:45:00] Phase 1 complete: 01-spec.md
[2026-04-06T14:45:30] User approved spec
[2026-04-06T14:50:00] Codex review spec started (job-id: task-abc123)
[2026-04-06T14:55:00] Codex review spec complete: 02-codex-review-spec.md
[2026-04-06T14:55:30] Decision #1: A (integrate all)
[2026-04-06T14:58:00] Spec corrected: 03-spec-v2.md
[2026-04-06T15:10:00] Phase 3 complete: 04-plan.md
[2026-04-06T15:15:00] Codex review plan started (job-id: task-def456)
[2026-04-06T15:22:00] Codex review plan complete: 05-codex-review-plan.md
[2026-04-06T15:22:30] Decision #2: B (integrate findings 1,3, ignore 2)
[2026-04-06T15:25:00] Plan corrected: 06-plan-v2.md
[2026-04-06T15:26:00] Phase 5 started (base_ref: a1b2c3d)
[2026-04-06T15:45:00] Phase 5 complete: 5 commits (a1b..f6e)
[2026-04-06T15:50:00] Codex review impl started (job-id: task-ghi789)
[2026-04-06T16:02:00] Codex review impl complete: 07-codex-review-impl.md
[2026-04-06T16:02:30] Decision #3: A (fix all)
[2026-04-06T16:10:00] Phase 7: VERIFICATION_PASSED
This phase runs automatically if a project wiki exists (wiki/schema.md).
If no wiki exists, skip silently. The user can set one up later with /wiki init.
The pipeline session is a rich knowledge source — decisions, trade-offs, reviews, patterns. Auto-ingest the session as a single source:
wiki/schema.md exists. If not → skip Phase 8 entirely.wiki/schema.md and wiki/index.md for context.wiki/sources/YYYY-MM-DD-pipeline-<slug>.md:
# Pipeline Session: <slug>
## Spec Summary
[Key decisions and trade-offs from 01-spec.md or 03-spec-v2.md]
## Codex Findings
[Important findings from reviews — bugs caught, patterns flagged]
## Implementation Decisions
[Architecture choices, rejected alternatives, lessons learned]
## Verification
[What was tested, what passed, what needed fixing]
wiki/pages/sources/YYYY-MM-DD-pipeline-<slug>.md.wiki/index.md with new/modified pages.wiki/log.md:
## [YYYY-MM-DD HH:MM] INGEST | Pipeline session: <slug>
- Source: sources/YYYY-MM-DD-pipeline-<slug>.md
- Pages created: <list>
- Pages updated: <list>
Wiki updated: +N pages from pipeline session
Log: [timestamp] Phase 8 complete: wiki updated (N pages)
docs/superpowers/specs/, docs/superpowers/plans/). CLoClo copies or symlinks them into the session dir for session tracking.npx claudepluginhub bacoco/cloclo --plugin clocloProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.