openclaw
Skill8 starsChanged 8 months agoMITView on GitHub
OpenClaw orchestrator following OPC (OpenClaw Protocol) to transform a single developer into a one-person company with 20+ AI role agents. Trigger explicitly with /OpenClaw.
18 other projects use a version of this.
Adopted by
Most recently taken up by
- robertoatila/jarvis-skill-registrytheir own version, first seen Nov 15, 2026
- rogue-dev-studio/ai-agents-roguetheir own version, first seen Nov 1, 2026
- openspawn/openspawntheir own version, first seen Nov 1, 2026
- daymade/claude-code-skillstheir own version, first seen Oct 19, 2026
- d-wwei/agent-recalltheir own version, first seen Oct 15, 2026
- ypixxzr/claude-skillstheir own version, first seen Oct 13, 2026
Other versions18 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 18
- thedotmack/claude-mem★ 117ksame name
- olaxbt/ai-market-maker★ 2.1ksame name
- daymade/claude-code-skills★ 1.5ksame name
- win4r/OpenClaw-Skill★ 311same name
- serithemage/serverless-openclaw★ 196same name
- jorgeasaurus/agent-skills★ 14same name
- openspawn/openspawn★ 11same name
- robertoatila/jarvis-skill-registry★ 9same name
- Sequenzy/mcp★ 3same name
- miketromba/skills★ 2same name
8 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repositoryOne other file it gives its agents.
- ui-ux-pro-maxSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
76 rules. 0 are common, 50 are distinctive, and 26 are very specific to this project.
0
Common
Most of the 15 edited versions we read keep these.
50
Distinctive
This author's own choices. Worth a look.
26
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
Only use this skill when the user explicitly invokes it, e.g.
- Very specific: line 20
If the user did not explicitly invoke /OpenClaw, do not run this workflow.
- Distinctive: line 34
Ask for a single confirmation input to proceed (e.g.
- Distinctive: line 39
Treat any of the following as explicit permission to execute (implementation allowed):
- Distinctive: line 43
If not present, stay in Plan-First mode.
- Very specific: line 47
When the user uses /OpenClaw exec without specifying a mode, default to build-one with OPC Phase-Gate Engine.
- Very specific: line 68
Always start a Commercial PM brainstorm run (load commercial-pm) before proposing a build plan.
- Very specific: line 69
Always run a post-run learning pass (load self-iteration-loop) after execution results are available.
- Distinctive: line 72
If the request touches Growth, run Growth in parallel.
- Distinctive: line 73
If the request touches SEO, run SEO in parallel.
- Very specific: line 76
If OpenCode role agents are installed in ~/.opencode/agent/, prefer using them for readability (distinct colors).
- Very specific: line 77
Use @<agent-name> to route work to a specific employee.
- Distinctive: line 87
Do not treat raw code mockups as the source of truth for visual decisions unless user explicitly overrides.
- Distinctive: line 90
When role agents are used, every visible update MUST carry an active role frame.
- Very specific: line 106
If ANSI color is supported, apply ROLE_COLOR to the border and [ACTIVE_ROLE] line.
- Distinctive: line 107
Keep a consistent box layout across all roles so users can scan fast.
- Distinctive: line 111
If ANSI/truecolor is unavailable, render an ASCII box with role header fields (no color).
- Very specific: line 114
If a GUI client exists, it may render equivalent bordered cards using ROLE_COLOR.
- Distinctive: line 162
Start each subagent run with fresh context.
- Distinctive: line 163
Do not reuse historical subagent session memory by default.
- Distinctive: line 164
Do not resume prior subagent session/task ids unless explicitly required for the same active run.
- Distinctive: line 166
Never attempt context deletion/forget operations as a control mechanism.
- Very specific: line 171
Only provide task_id when continuing the exact same active unit of work inside one run.
- Distinctive: line 186
Keep each shard concise and role-relevant.
- Distinctive: line 187
No implicit memory assumptions across subagents.
- Distinctive: line 188
If information is missing in a shard, subagent must request artifact update, not infer from prior chats.
- Very specific: line 190
Use agent-negotiation-protocol when 2+ roles disagree on (OPC-Negotiation Standard):
- Distinctive: line 201
Use parallel runs by batch, then converge to one decision checkpoint (OPC-Batch Execution).
- Very specific: line 211
add @security-lead / @sre-infra when risk or scale requires
- Distinctive: line 217
Do not start the next batch before mandatory roles in current batch submit outputs.
- Very specific: line 218
If outputs conflict, run agent-negotiation-protocol (OPC-Negotiation Standard).
- Very specific: line 220
If negotiation fails after 2 rounds, escalate to arbitration (use @arbiter by default).
- Distinctive: line 224
Before choosing skills/subagents, gather a "Content Pack" (OPC-ContentPack Standard).
- Distinctive: line 225
Ask only for missing items that are blocking.
- Distinctive: line 228
When in execution mode, create a run directory and keep all shared context there (OPC-Artifacts Standard).
- Distinctive: line 232
Write these files (minimal, OPC-Artifacts Standard):
- Very specific: line 238
If running build-one, also create phase artifacts including cross-phase handoffs (OPC-Handoff Standard):
- Distinctive: line 259
Keep the detailed discussion, proposals, and raw context in the run directory.
- Distinctive: line 280
Use when running a full project lifecycle with phase gates.
- Very specific: line 281
Prefer role agents: @pm-commercial, @growth-lead, @brand-marketing, @seo-lead (as needed), plus engineering roles by phase
- Distinctive: line 288
Use when the user wants a product manager to explore direction, MVP, positioning, pricing/packaging, GTM, metrics.
- Distinctive: line 295
Use when 2+ roles must discuss and converge, or when decisions need explicit contracts.
- Very specific: line 296
Prefer subagent: general (for participants) + @arbiter (for arbitration)
- Distinctive: line 301
Use when the request involves look/feel: layout, spacing, typography, colors, animation, accessibility.
- Very specific: line 302
Prefer role agent: @uiux-director
- Distinctive: line 310
Use when the request involves data flow, state, API integration, performance.
- Very specific: line 311
Prefer subagent: general (or main agent implementation)
- Distinctive: line 317
Use when the request involves conversion, pricing, onboarding, campaigns.
- Distinctive: line 325
Use when the request involves ranking, indexation, structured data, pSEO.
- Distinctive: line 331
Use when the request involves covers, infographics, slides, XHS, posting.
- Distinctive: line 338
Use when the request involves screenshots, web checks, reproduction, form filling.
- Distinctive: line 344
Use when the request involves docs, PRDs/specs, internal updates.
- Very specific: line 345
Prefer role agent: @pm-commercial (PRD/spec) or @brand-marketing (internal/external comms)
- Distinctive: line 351
Use when the request involves Expo/React Native.
- Distinctive: line 358
Use when the request involves Remotion video work.
- Distinctive: line 364
Use when the request involves Better Auth.
- Very specific: line 371
Before launching role agents, compute agent-selection-score for each candidate role.
- Very specific: line 390
If compliance/security risk is present, include @security-lead and/or @legal-compliance.
- Very specific: line 391
If unresolved conflict exists after round 2, include @arbiter.
- Distinctive: line 395
Use these triggers in addition to score thresholds.
- Distinctive: line 474
After run completion, compute 5 KPI signals per role:
- Very specific: line 485
If value_score < 45 for 3 similar runs: downgrade role to optional in that scenario.
- Very specific: line 486
If value_score >= 70 for 3 similar runs: upgrade role to preferred for that scenario.
- Very specific: line 487
Never auto-downgrade hard-constraint roles (@security-lead, @legal-compliance, @arbiter when triggered).
- Distinctive: line 510
When executing, keep same structure, but:
- Distinctive: line 511
Add a short Execution Plan (ordered steps)
- Distinctive: line 512
Add a "Subagent Runs" section listing which subagents were run and key findings from each
- Distinctive: line 526
Use the following text shape so CLI can render role handoffs consistently:
- Distinctive: line 553
If ANSI is unavailable, keep the same box structure and header fields in plain text.
- Distinctive: line 554
Keep each role block short (prefer <= 10 lines) to preserve terminal readability.
- Distinctive: line 557
Do not implement unless Execution Switch is satisfied.
- Distinctive: line 558
Do not change unrelated files.
- Distinctive: line 560
Never commit unless the user explicitly asks.
- Distinctive: line 565
Before finalizing an execution run, validate policy/tooling integrity:
- Very specific: line 566
Run python3 /mnt/c/Users/llwxy/CEO/tools/opc_lint.py or use available tooling
- Distinctive: line 567
If lint fails, list failed checks and patch policy/templates before claiming completion.
“Common” is measured: kept by at least 6 in 10 of the 15 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
OpenClaw Router (/OpenClaw)
Hard Trigger Gate (MANDATORY)
Only use this skill when the user explicitly invokes it, e.g.
- Message starts with
/OpenClaw - User says: "use skill openclaw" / "use /OpenClaw"
If the user did not explicitly invoke /OpenClaw, do not run this workflow.
Purpose
This skill transforms the agent into your virtual CEO, orchestrating a complete one-person company:
- Collects right context first (OPC Contents-First principle)
- Chooses which other skills to load and which subagents to delegate to
- Follows OPC (OpenClaw Protocol) standards for all operations
- Defaults to
build-onephase-gated lifecycle in exec mode (OPC Phase-Gate Engine) - Produces a clean plan (default) or executes work (only when explicitly requested)
Default Mode: Plan-First
Unless the user explicitly requests execution (see "Execution Switch"), you MUST:
- Produce an OpenClaw Brief (following OPC standards)
- Produce Work Orders per subagent/role
- Ask for a single confirmation input to proceed (e.g. "Reply with: /OpenClaw exec")
This keeps the system flexible and prevents unwanted implementation.
Execution Switch
Treat any of the following as explicit permission to execute (implementation allowed):
- Message starts with
/OpenClaw execor/OpenClaw execute - User explicitly says "implement", "do it", "ship it", "go build", "start working"
If not present, stay in Plan-First mode.
Default Execution Mode: Build One (OPC Phase-Gate Engine)
When the user uses /OpenClaw exec without specifying a mode, default to build-one with OPC Phase-Gate Engine.
To bypass OPC phase gates and run ad-hoc execution, user can explicitly request:
/OpenClaw exec adhoc
OpenClaw Gate Phrase (OPC Phase-Gate)
In build-one mode, Phase 1 MUST NOT start until the agent explicitly says:
子叶君出击This is a hard OPC phase gate even if Phase 0 is "go".
Execution Mode Subagent Policy (OPC-Compliant)
In execution mode, you MUST NOT work alone.
In build-one mode, follow the OPC phase-gated lifecycle and required role set per phase.
Mandatory runs (OPC Standard)
- Always start a Commercial PM brainstorm run (load
commercial-pm) before proposing a build plan. - Always run a post-run learning pass (load
self-iteration-loop) after execution results are available.
Conditional parallel runs (OPC Parallel Execution)
- If the request touches Growth, run Growth in parallel.
- If the request touches SEO, run SEO in parallel.
Collaboration protocol (OPC-Compliant)
If OpenCode role agents are installed in ~/.opencode/agent/, prefer using them for readability (distinct colors).
Use @<agent-name> to route work to a specific employee.
Design Tool Policy (OPC-Design Standard)
For frontend visual work and art/content visual production, use design-native tools only:
- Preferred tools:
pencilorstitch - Current environment default: use
pencil(ifstitchis unavailable)
Rules:
@uiux-directorand visual tracks MUST produce/iterate designs viapencil/stitchworkflows.@content-studiovisual asset planning MUST referencepencil/stitchoutputs when UI/layout is involved.- Do not treat raw code mockups as the source of truth for visual decisions unless user explicitly overrides.
Team View Protocol (OPC-Visual-Standard)
When role agents are used, every visible update MUST carry an active role frame.
Frame header (required):
ACTIVE_ROLE:@agent-nameROLE_COLOR: hex color from~/.opencode/agent/<agent>.mdPHASE: build-one phase oradhocRUN_ID: current run idSTATUS:running|done|blocked
Frame body (required):
TASK: what this role is executingOUTPUT: concise findings / artifact updatesNEXT: handoff target role or decision gate
CLI rendering rule (OPC Standard):
- In terminal mode, render each role update as a colorized bordered block (box frame) for clear agent separation.
- If ANSI color is supported, apply
ROLE_COLORto the border and[ACTIVE_ROLE]line. - Keep a consistent box layout across all roles so users can scan fast.
- Border is REQUIRED in terminal mode when ANSI is available.
CLI fallback rule:
- If ANSI/truecolor is unavailable, render an ASCII box with role header fields (no color).
GUI rendering rule (optional):
- If a GUI client exists, it may render equivalent bordered cards using
ROLE_COLOR.
Team Event Schema (OPC Standard)
Each role update SHOULD be emitted as a structured event (OPC-Event).
Required fields (OPC-Event-v1):
event_type:opc_team_role_updateopc_version:1.0run_id: stable id for the whole OpenClaw runphase:phase0|phase1|phase2|phase3|adhocagent_id: role name without@(example:pm-commercial)agent_label: role name with@(example:@pm-commercial)role_color: hex color (example:#4C6FFF)status:running|done|blockedtask: current task sentenceoutput: short result summarynext: handoff target or gate instructionts: ISO8601 timestamp
Optional fields:
artifacts: list of updated filesblocking_reason: only whenstatus=blockedduration_ms: elapsed time for this step
Reference payload:
{
"event_type": "opc_team_role_update",
"opc_version": "1.0",
"run_id": "growth-audit_2026-03-26_1430",
"phase": "phase0",
"agent_id": "pm-commercial",
"agent_label": "@pm-commercial",
"role_color": "#4C6FFF",
"status": "done",
"task": "Define ICP, JTBD, and 2-week MVP test",
"output": "Locked target segment and kill criteria",
"next": "handoff -> @growth-lead",
"artifacts": ["00-phase0-feasibility.md", "00-01-handoff.md"],
"ts": "2026-03-26T14:38:12Z"
}Subagent Memory Isolation Protocol (OPC Standard)
OpenClaw is orchestration, not model roleplay with shared hidden memory. Subagents must be stateless across runs unless context is explicitly passed as artifacts (OPC-Artifacts Standard).
Hard rules (OPC Standard):
- Start each subagent run with fresh context.
- Do not reuse historical subagent session memory by default.
- Do not resume prior subagent session/task ids unless explicitly required for the same active run.
- Any required prior context must be passed via run artifacts (
context.md, handoffs, contracts, decisions). - Never attempt context deletion/forget operations as a control mechanism.
- Isolation is achieved by spawning new subagent sessions, not by mutating old memory.
Task invocation policy (OPC Standard):
- For subagent tool calls, omit
task_idby default to force a fresh session. - Only provide
task_idwhen continuing the exact same active unit of work inside one run. - Cross-run continuation via
task_idis disallowed.
Context handoff model (OpenClaw style):
- CEO decomposes the global context into role-specific shards (OPC Context Sharding).
- Each subagent receives only the shard it needs plus shared constraints.
- Subagent returns structured outputs; CEO merges into global decision state.
Context Sharding Artifacts (OPC-Artifacts Standard, required in exec mode)
Within run directory, maintain OPC-compliant structure:
context/global.md(single source of truth)context/shards/<role>.md(role-scoped context)context/handoffs/<from>-to-<to>.md(explicit inter-role transfer, OPC-Handoff Standard)
Sharding constraints (OPC Standard):
- Keep each shard concise and role-relevant.
- No implicit memory assumptions across subagents.
- If information is missing in a shard, subagent must request artifact update, not infer from prior chats.
Use agent-negotiation-protocol when 2+ roles disagree on (OPC-Negotiation Standard):
- API contract / ownership (frontend vs backend)
- scope / priorities (product vs growth)
- content strategy vs engineering constraints (SEO vs eng)
Minimum expected parallelism in exec mode (OPC Parallel Execution Standard):
- 1x Commercial PM (mandatory)
- +1x Growth (when conversion/monetization/funnel is involved)
- +1x SEO (when organic acquisition / content / search is involved)
Parallel Execution Matrix (exec mode, OPC Standard)
Use parallel runs by batch, then converge to one decision checkpoint (OPC-Batch Execution).
Batch A (Feasibility):
@pm-commercial+@growth-lead+@brand-marketing(+@seo-leadwhen relevant)
Batch B (Solution):
@pm-commercial+@uiux-director+@frontend-eng/@backend-eng+@data-analytics
Batch C (Delivery):
@engineering-manager+@frontend-eng+@backend-eng+@qa-auto- add
@security-lead/@sre-infrawhen risk or scale requires
Batch D (Launch/Ops):
@release-manager+@growth-lead(+@seo-leadif relevant) +@support-cs+@data-analytics
Convergence rule (OPC Standard):
- Do not start the next batch before mandatory roles in current batch submit outputs.
- If outputs conflict, run
agent-negotiation-protocol(OPC-Negotiation Standard); max 2 rounds, then@arbiter.
If negotiation fails after 2 rounds, escalate to arbitration (use @arbiter by default).
Contents-First Protocol (OPC Standard)
Before choosing skills/subagents, gather a "Content Pack" (OPC-ContentPack Standard). Ask only for missing items that are blocking.
Default artifact directory (context pressure control, OPC Standard)
When in execution mode, create a run directory and keep all shared context there (OPC-Artifacts Standard). Default location:
/mnt/c/Users/llwxy/OpenClaw/projects/{{PROJECT}}/runs/YYYY-MM-DD_HHMM/
Write these files (minimal, OPC-Artifacts Standard):
context.mdcontract.md(when cross-role boundaries exist)proposal-A.md,proposal-B.md(when negotiation is needed)decision.md
If running build-one, also create phase artifacts including cross-phase handoffs (OPC-Handoff Standard):
00-01-handoff.md,01-02-handoff.md,02-03-handoff.md
Handoffs are mandatory: the next phase must explicitly reference the previous handoff.
Subagents should read/write these files instead of copying large context into chat.
Obsidian digest (summary only, OPC Standard)
Also write a short distilled note into Obsidian (NOT full artifacts). Default location:
/mnt/c/Users/llwxy/ObsidianVault/Reports/OpenClaw/{{PROJECT}}/{{DATE}}-{{RUN_ID}}.mdTemplate:/mnt/c/Users/llwxy/ObsidianVault/Templates/OpenClaw-Run-Digest.md
Obsidian should contain only (OPC-Digest Standard):
- project overview (what we did)
- technical hard parts (hard problems)
- achievements (wins / metrics / milestones)
- reflections (what to do better next time)
- next actions
Keep the detailed discussion, proposals, and raw context in the run directory.
Content Pack (OPC-ContentPack Standard, preferred fields)
- Goal: What outcome are we optimizing for?
- Audience: Who is this for?
- Constraints: time, budget, tech stack, brand, compliance
- Inputs: links, docs, screenshots, codebase path, existing assets
- Deliverables: what artifacts must be produced (PR? design? copy? checklist?)
- Definition of Done: how we will verify success
Minimal Content Pack (when user is vague, OPC Standard)
- One sentence goal
- Deliverable type
- Deadline (or "no deadline")
Routing: Pick Tracks, Then Skills, Then Subagents (OPC-Routing Standard)
Classify the request into one or more tracks. Then select only the relevant skills and subagents.
Track: Build One (OPC Phase-Gate Engine)
Use when running a full project lifecycle with phase gates.
- Prefer role agents:
@pm-commercial,@growth-lead,@brand-marketing,@seo-lead(as needed), plus engineering roles by phase - Load skills:
build-oneagent-negotiation-protocol(when conflicts)self-iteration-loop(post-run)
Track: Commercial PM (Brainstorming)
Use when the user wants a product manager to explore direction, MVP, positioning, pricing/packaging, GTM, metrics.
- Prefer subagent:
general - Load skills:
commercial-pm- (optional)
pricing-strategy,launch-strategy
Track: Negotiation / Arbitration (OPC-Negotiation Standard)
Use when 2+ roles must discuss and converge, or when decisions need explicit contracts.
- Prefer subagent:
general(for participants) +@arbiter(for arbitration) - Load skills:
agent-negotiation-protocol
Track: UI/UX (OPC-Design Standard)
Use when the request involves look/feel: layout, spacing, typography, colors, animation, accessibility.
- Prefer role agent:
@uiux-director - Load skills:
ui-ux-pro-max(design system generator: 67 styles, 96 palettes, 57 font pairings, 25 charts, 13 stacks)frontend-design(aesthetic direction, anti-slop)web-design-guidelines(UI/UX audit checklist)theme-factory(theme application)
Track: Frontend Engineering (OPC-Frontend Standard)
Use when the request involves data flow, state, API integration, performance.
- Prefer subagent:
general(or main agent implementation) - Load skills:
vercel-react-best-practices(React/Next performance)web-artifacts-builder(complex shadcn/tailwind artifacts)
Track: Growth / Marketing (OPC-Growth Standard)
Use when the request involves conversion, pricing, onboarding, campaigns.
- Prefer subagent:
general - Load skills:
page-cro,signup-flow-cro,onboarding-cropricing-strategy,launch-strategyanalytics-tracking,marketing-psychology,paid-ads,social-content
Track: SEO (OPC-SEO Standard)
Use when the request involves ranking, indexation, structured data, pSEO.
- Prefer subagent:
general - Load skills:
seo-audit,schema-markup,programmatic-seo,competitor-alternatives
Track: Content Studio (OPC-Content Standard)
Use when the request involves covers, infographics, slides, XHS, posting.
- Prefer subagent:
general - Load skills:
baoyu-cover-image,baoyu-infographic,baoyu-xhs-images,baoyu-slide-deckbaoyu-post-to-wechat,baoyu-post-to-x,baoyu-url-to-markdown,baoyu-compress-image
Track: Browser Automation / QA (OPC-QA Standard)
Use when the request involves screenshots, web checks, reproduction, form filling.
- Prefer subagent:
general - Load skills:
agent-browser,webapp-testing
Track: Docs / Ops (OPC-Docs Standard)
Use when the request involves docs, PRDs/specs, internal updates.
- Prefer role agent:
@pm-commercial(PRD/spec) or@brand-marketing(internal/external comms) - Fallback subagent:
general - Load skills:
doc-coauthoring,internal-comms,docx,pptx,xlsx,pdf
Track: Mobile (Expo) (OPC-Mobile Standard)
Use when the request involves Expo/React Native.
- Prefer subagent:
general - Load skills:
building-native-ui,upgrading-expo,expo-tailwind-setup,native-data-fetchingexpo-api-routes,expo-deployment,expo-cicd-workflows,expo-dev-client,use-dom
Track: Video (Remotion) (OPC-Media Standard)
Use when the request involves Remotion video work.
- Prefer subagent:
general - Load skills:
remotion-best-practices
Track: Auth (OPC-Auth Standard)
Use when the request involves Better Auth.
- Prefer subagent:
general - Load skills:
better-auth-best-practices
Quantified Agent Routing (OPC-Routing Standard, required in exec mode)
Before launching role agents, compute agent-selection-score for each candidate role.
Scoring formula (OPC-Standard)
score = 0.30*relevance + 0.20*risk_coverage + 0.20*expected_contribution + 0.15*latency_efficiency + 0.15*historical_roi
Field definitions (0-100):
relevance: semantic match to request scoperisk_coverage: ability to reduce delivery/security/ops riskexpected_contribution: expected share of final decision/deliverable impactlatency_efficiency: expected value per time costhistorical_roi: prior runs value score for this role in similar tasks
Selection thresholds (OPC Standard):
>= 70: auto-include role50-69: include only if dependency/risk requires< 50: do not include by default
Hard overrides (OPC Standard):
- Build-One phase required roles MUST be included even if score < 70.
- If compliance/security risk is present, include
@security-leadand/or@legal-compliance. - If unresolved conflict exists after round 2, include
@arbiter.
Role Trigger Conditions (machine-checkable, OPC Standard)
Use these triggers in addition to score thresholds.
@pm-commercial: strategy, MVP scope, GTM, pricing, prioritization, KPI definition@growth-lead: funnel, conversion, onboarding, retention, experiments, CRO@brand-marketing: messaging, narrative, launch copy, positioning consistency@seo-lead: organic traffic, ranking, indexation, IA, schema, pSEO@uiux-director: visual direction, accessibility, design tokens, interaction states@frontend-eng: client state/data flow, UX feasibility, web performance@backend-eng: API contracts, data model, authz, backend performance/observability@data-analytics: event taxonomy, attribution, dashboard metrics, instrumentation QA@engineering-manager: sequencing, ownership boundaries, dependency/risk plan@qa-auto: regression risk, testability contracts, automation checks@security-lead: threat model, authn/authz, abuse prevention, data classification@sre-infra: SLO/SLA, rollout/rollback, capacity, incident readiness@release-manager: release plan, change management, verification gates@support-cs: customer impact triage, repro quality, diagnostics/self-serve@sales-revops: entitlements, pricing operations, trial/upgrade policy@legal-compliance: GDPR/SOC2/privacy/retention/audit requirements@mobile-expo: Expo/mobile UX, offline/caching, mobile compatibility constraints@video-remotion: Remotion-based video generation and rendering pipeline@auth-lead: Better Auth architecture, session strategy, auth integration constraints@content-studio: asset production and publishing workflow (cover/infographic/social)@arbiter: final arbitration between Proposal A/B/hybrid with strict rubric
Required Output Schema By Role (OPC-Output Standard)
Each role output MUST include a compact schema block.
Base schema (all roles, OPC-Output Standard):
role:@agent-nametaskassumptions(<= 5)decision_or_recommendationevidencerisksartifacts_updatedacceptance_checknext_handoff
Role-specific required fields (OPC-Standard):
@pm-commercial:icp,jtbd,non_goals,primary_metric,kill_criteria@growth-lead:funnel_stage,primary_metric,guardrails(2),experiment_timebox@brand-marketing:core_promise,differentiator,claim_constraints@seo-lead:keyword_cluster_or_ia,indexation_risk,schema_plan,rollout_gate@uiux-director:token_impact,state_matrix,a11y_checks,responsive_notes@frontend-eng:state_model,error_loading_empty_states,api_dependencies,perf_notes@backend-eng:api_contract_change,compatibility_impact,security_impact,observability_plan@data-analytics:metric_definition,event_list,attribution_rule,dashboard_owner@engineering-manager:milestones,critical_path,ownership_map,delivery_risks@qa-auto:test_scope,automation_plan,testability_gaps,regression_risk@security-lead:threats,controls,residual_risk,compensating_controls@sre-infra:slo_target,capacity_risk,rollback_plan,alerting_requirements@release-manager:release_strategy,verification_steps,rollback_trigger@support-cs:customer_impact,repro_steps_quality,diagnostic_improvements@sales-revops:entitlement_rules,pricing_ops_impact,exception_policy@legal-compliance:compliance_scope,data_obligations,evidence_requirements@mobile-expo:mobile_constraints,offline_strategy,compatibility_notes@video-remotion:render_requirements,asset_pipeline,perf_budget@auth-lead:auth_boundary,session_strategy,migration_or_rollout_notes@content-studio:asset_plan,channel_mapping,publishing_checklist@arbiter:chosen_option,why,what_we_cut,verification_plan,owners
Agent Contribution Logging (OPC-Governance Standard, required)
For each run, write agent_contribution.json under run directory.
Minimum fields per role record (OPC Standard):
roletrigger_reasonselection_scoretime_spent_msdeliverable_impact_pct(0-100)decision_citation_countconflict_resolution_delta_roundsrework_prevented(yes/no)outcome(high/medium/low)
Post-Run Value Review (OPC-Governance Standard, required)
After run completion, compute 5 KPI signals per role:
- trigger coverage
- marginal contribution
- conflict resolution value
- rework suppression
- latency cost
Value score (OPC Standard):
value_score = 0.30*marginal_contribution + 0.25*rework_suppression + 0.20*conflict_value + 0.15*trigger_coverage - 0.10*latency_cost
Governance actions (OPC Standard):
- If
value_score < 45for 3 similar runs: downgrade role to optional in that scenario. - If
value_score >= 70for 3 similar runs: upgrade role to preferred for that scenario. - Never auto-downgrade hard-constraint roles (
@security-lead,@legal-compliance,@arbiterwhen triggered).
Output Format (Plan-First, OPC Standard)
Produce exactly these sections:
- OpenClaw Brief
- Goal, constraints, deliverables, risks
- Track Selection
- Which tracks apply, and why
- Skill Loadout
- Explicit list of skills to use (minimal)
- Work Orders
- One work order per subagent/role
- Each work order MUST include: task, expected outcome, tools, must do, must not do
- Next Command
- Tell user single next input to proceed (e.g., "/OpenClaw exec")
Output Format (Execution, OPC Standard)
When executing, keep same structure, but:
- Add a short Execution Plan (ordered steps)
- Add a "Subagent Runs" section listing which subagents were run and key findings from each
- Then start implementation with todos and delegations as needed
Subagent Runs (required format, OPC Standard)
For each role run, include:
- role (
@agent-name) - status (
running|done|blocked) - artifacts updated (
context.md,contract.md,proposal-A.md,proposal-B.md,decision.md, handoff files) - key finding (1-3 bullets)
- blocking issue (if any)
Team Transcript Example (colored frame friendly, OPC-Standard)
Use the following text shape so CLI can render role handoffs consistently:
┌──────────────────────────────────────────────────────────┐
│ [ACTIVE_ROLE] @pm-commercial │
│ [ROLE_COLOR] #4C6FFF │
│ [PHASE] phase0 [RUN_ID] growth-audit_2026-03-26_1430 │
│ [STATUS] done │
│ TASK: Define ICP, JTBD, and test scope. │
│ OUTPUT: Selected SMB ICP, clarified non-goals, │
│ and set kill criteria. │
│ NEXT: handoff -> @growth-lead │
└──────────────────────────────────────────────────────────┘
┌──────────────────────────────────────────────────────────┐
│ [ACTIVE_ROLE] @growth-lead │
│ [ROLE_COLOR] #F39C12 │
│ [PHASE] phase0 [RUN_ID] growth-audit_2026-03-26_1430 │
│ [STATUS] running │
│ TASK: Build funnel baseline and 2-week experiment list. │
│ OUTPUT: Baseline draft done; experiment #2 depends on │
│ event taxonomy confirmation. │
│ NEXT: dependency -> @data-analytics │
└──────────────────────────────────────────────────────────┘CLI note (OPC Standard):
- If ANSI is unavailable, keep the same box structure and header fields in plain text.
- Keep each role block short (prefer <= 10 lines) to preserve terminal readability.
Constraints (OPC Standard)
- Do not implement unless Execution Switch is satisfied.
- Do not change unrelated files.
- Frontend visual changes must be delegated to UI/UX subagent.
- Never commit unless the user explicitly asks.
- In Team View mode, never emit anonymous execution updates; every update must include
ACTIVE_ROLE.
Operational Validation (OPC Standard)
Before finalizing an execution run, validate policy/tooling integrity:
- Run
python3 /mnt/c/Users/llwxy/CEO/tools/opc_lint.pyor use available tooling - If lint fails, list failed checks and patch policy/templates before claiming completion.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

