loop-workflow-planner
Skill7 starsChanged 21 days agoApache-2.0View on GitHub
Define a platform-neutral bounded loop for standard or deep work using measurable acceptance criteria, verification, retry limits, independent review, memory, and stop conditions. Skip lookups, tiny edits, and one-off explanations.
In this repository23 other files it gives its agents.
- AGENTS.mdAGENTS.md
- agent-rules-skill-authorSkill
- agent-self-improvementSkill
- design-screenshot-specSkill
- execution-plan-managerSkill
- frontend-architecture-plannerSkill
- frontend-bugfix-debuggerSkill
- frontend-design-directorSkill
- frontend-design-intelligenceSkill
- frontend-layout-implementerSkill
- frontend-linter-managerSkill
- frontend-prototype-explorerSkill
- frontend-quality-reviewerSkill
- frontend-refactor-surgeonSkill
- frontend-visual-qaSkill
- goal-plannerSkill
- greenfield-project-builderSkill
- mcp-toolchain-managerSkill
- pattern-library-managerSkill
- project-context-adapterSkill
- project-onboarding-adapterSkill
- webdev-feedbackSkill
- webdev-kit-updaterSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
19 rules. About 0 are common, 8 are distinctive, and 11 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
Distinctive · estimate
This author's own choices. Worth a look.
11
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 38
Create a platform-neutral Loop Workflow Contract for standard or deep work that must iterate until measurable acceptance criteria pass.
- Very specific: line 44
Use this skill when the task is Standard Workflow or Deep Workflow and one or more are true:
- Distinctive: line 56
Do not use for:
- Very specific: line 73
Read common/context-compaction-rules.md when stop/resume or memory matters.
- Very specific: line 74
Read common/independent-review-rules.md when review is needed.
- Very specific: line 75
Read common/worktree-parallelism-rules.md only when parallel agents, branches, or worktrees are considered.
- Very specific: line 76
Read the current Goal Contract, Execution Plan, user request, or project/active-goals.md when present.
- Very specific: line 77
Read the active coverage map and target S-### slices when a durable plan exists.
- Very specific: line 78
Read project/verification-profile.md when verification commands matter.
- Very specific: line 80
Do not read unrelated skills or generated dist/ during normal runtime.
- Distinctive: line 87
Must not implement source code.
- Distinctive: line 88
Must not run verification commands.
- Distinctive: line 89
Must not install packages, tools, MCP servers, UI libraries, or testing workflows.
- Distinctive: line 90
Must not create parallel worktrees or branches without explicit user approval and an ownership plan.
- Distinctive: line 100
Do not add, reorder, renumber, or mark slices from the loop contract.
- Very specific: line 140
When a Goal Contract exists, preserve its AC-### identifiers in the loop contract, verification evidence, retry history, and final report.
- Very specific: line 142
When an Execution Plan exists, preserve its S-### identifiers in target scope, retry history, memory, review handoff, and resume state.
- Very specific: line 145
Use templates/loop-workflow-contract.md for durable loop contracts.
- Distinctive: line 153
Every durable retry attempt must identify its affected criteria and slices.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Loop Workflow Planner
Purpose
Create a platform-neutral Loop Workflow Contract for standard or deep work that must iterate until measurable acceptance criteria pass.
This skill does not implement code. It defines objective, scope, verification, retry limits, stop conditions, independent review, and memory update rules so Codex, GPT-based agents, Claude, and generic coding agents can execute the same workflow without depending on Claude-only commands.
When To Use
Use this skill when the task is Standard Workflow or Deep Workflow and one or more are true:
- the user asks to continue until lint, typecheck, build, visual QA, CI, or another measurable check passes;
- verification may fail and repair is in scope;
- a repeated failure already happened;
- a bugfix, CI repair, refactor, or UI implementation needs bounded retry;
- independent review should judge completion;
- the work may need resumable memory;
- multiple agents or worktrees are being considered.
When Not To Use
Do not use for:
- Fast Lookup;
- one obvious typo or type error;
- tiny direct edits;
- pure explanation;
- design-only direction with no iterative verification;
- platform-specific
/goal,/loop, or/schedulecommands without a platform-neutral contract; - unbounded autonomous work.
Required Context
- Read
AGENTS.md. - Read
common/agent-loop-policy.md. - Read
common/stop-criteria-rules.md. - Read
common/bounded-retry-rules.md. - Read
common/verification-loop-rules.md. - Read
common/context-compaction-rules.mdwhen stop/resume or memory matters. - Read
common/independent-review-rules.mdwhen review is needed. - Read
common/worktree-parallelism-rules.mdonly when parallel agents, branches, or worktrees are considered. - Read the current Goal Contract, Execution Plan, user request, or
project/active-goals.mdwhen present. - Read the active coverage map and target
S-###slices when a durable plan exists. - Read
project/verification-profile.mdwhen verification commands matter.
Do not read unrelated skills or generated dist/** during normal runtime.
Tool Contract
- May read project overlays and verification profiles.
- May write or update a durable loop contract only when the task is genuinely resumable or deep.
- May recommend host mappings for Claude, Codex, GitHub, or generic agents, but must not require host-specific commands.
- Must not implement source code.
- Must not run verification commands.
- Must not install packages, tools, MCP servers, UI libraries, or testing workflows.
- Must not create parallel worktrees or branches without explicit user approval and an ownership plan.
Workflow
- Confirm task scale. Stop if the task is Fast Lookup or lightweight without repeated failure.
- Confirm or derive the objective from an existing goal, execution plan, or explicit user request.
- Reuse the active Goal Contract
AC-###criteria when present. Otherwise, convert vague completion language into measurable acceptance criteria without creating a competing set of criteria. - Reference the target
S-###slices when a durable execution plan exists. Do not add, reorder, renumber, or mark slices from the loop contract. - Define allowed scope and out-of-scope changes.
- Select loop type: one-pass verification, bounded retry, goal-based loop, or open exploration.
- Set maximum attempts or turns.
- Define verification commands, rendered checks, or manual evidence sources.
- Define retry strategy and what must change after a failed attempt. Every
durable retry record must name the affected
AC-###andS-###. - Define stop and escalation conditions.
- Decide whether independent review is required.
- Decide whether loop memory is needed and where it belongs.
- Decide whether parallel work is safe; default to no parallelism.
- Produce a Loop Workflow Contract.
- Hand off to the next smallest relevant skill.
Output Contract
Final response: return only facts that affect the user's understanding, confidence, or next action. Omit empty fields and workflow narration.
Return or write:
Loop Workflow Contract
Goal ID
Objective
Allowed Scope
Acceptance Criteria
Target Slices
Loop Type
Max Attempts Or Turns
Retry Strategy
Verification
Independent Review
Memory Update
Stop Conditions
Escalation Conditions
Final Evidence
Next Slice
Next Skill Or Next StepWhen a Goal Contract exists, preserve its AC-### identifiers in the loop
contract, verification evidence, retry history, and final report.
When an Execution Plan exists, preserve its S-### identifiers in target scope,
retry history, memory, review handoff, and resume state.
Use templates/loop-workflow-contract.md for durable loop contracts.
Validation Gates
- The loop must be bounded.
- Acceptance criteria must be measurable enough for an independent reviewer.
- Existing goal criterion identifiers must be reused instead of redefined.
- Existing slice identifiers must be referenced instead of redefined or mutated.
- Every durable retry attempt must identify its affected criteria and slices.
- Verification must rely on existing project commands or available rendered checks.
- Retry strategy must prevent repeating the same failed approach.
- Independent review must be defined when material risk exists.
- Memory updates must stay in local-only
project/**files. - The contract must be executable by Codex/GPT and Claude without requiring platform-specific loop commands.
Trigger Evals
Should trigger:
- "Fix CI and keep going until it is green, but stop if it needs approval."
- "Repair this visual QA issue and rerun the rendered check until it passes."
- "Refactor this safely in iterations and verify after each slice."
- "Create a loop plan for this multi-step frontend task."
- "Use a goal loop, but keep it compatible with Codex and Claude."
Should not trigger:
- "Where is this component defined?"
- "Explain this function."
- "Fix this typo."
- "Change this button label."
- "Create a visual direction only."
Reference Map
common/agent-loop-policy.mdcommon/stop-criteria-rules.mdcommon/bounded-retry-rules.mdcommon/verification-loop-rules.mdcommon/context-compaction-rules.mdcommon/independent-review-rules.mdcommon/worktree-parallelism-rules.mdtemplates/loop-workflow-contract.mdtemplates/loop-memory.mdproject/loop-memory.mdoptional local-only memory file
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.

