spec-mode
Skill0 starsChanged 28 days agoMITView on GitHub
ChatGPT WebUI specification work for projects without a native change workflow, plus a compatibility entry point that delegates OpenSpec projects to openspec-workflow.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- sdwurg180507280211/my-skillstheir own version, first seen Oct 13, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- sdwurg180507280211/my-skills★ 1same name
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 repository12 other files it gives its agents.
- anti-sloprushSkill
- dev-modeSkill
- game-conceptSkill
- geometric-designSkill
- intent-modeSkill
- openspec-workflowSkill
- plan-modeSkill
- seo-aiseoSkill
- video-watchSkill
- visual-chatSkill
- visual-gauntletSkill
- visual-prototypeSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
20 rules. About 0 are common, 17 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈17
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 12
If the target project already uses OpenSpec and this is a non-trivial code change, load openspec-workflow and delegate immediately.
- Distinctive: line 14
Do not require all product and architecture decisions to be settled before involving OpenSpec.
- Distinctive: line 29
If a material decision is genuinely unresolved, do not invent it.
- Very specific: line 35
If the target project already uses OpenSpec, this mode delegates to openspec-workflow.
- Distinctive: line 37
Do not recreate OpenSpec semantics inside Spec Mode.
- Distinctive: line 41
If the project does not already define a specification system, use the default OpenSpec-style shape:
- Distinctive: line 60
Keep acceptance evidence explicit.
- Distinctive: line 69
Do not turn implementation guesses into requirements.
- Distinctive: line 80
keep scope aligned with the Plan Mode handoff.
- Distinctive: line 82
Do not repeatedly ask the user to approve routine specification details that are derivable from settled decisions and project facts.
- Distinctive: line 94
If new evidence changes the implementation route without changing intent, update the design coherently.
- Distinctive: line 96
If new evidence changes the intended behavior, scope, or user-visible contract, stop that branch and request a planning decision instead of …
- Distinctive: line 116
Before declaring completion, perform a coherence pass across all artifacts.
- Distinctive: line 122
Keep the canonical spec in the target project's existing source of truth whenever writes are authorized.
- Distinctive: line 134
Do not create a parallel shadow spec merely to checkpoint progress.
- Distinctive: line 142
Treat 64 as an observed GitHub connector boundary, not a universal guarantee.
- Very specific: line 168
When the user sends continue, resume the exact unfinished operation.
- Distinctive: line 190
Do not ask the user to restate information that exists in the repository or durable artifacts.
- Distinctive: line 194
When ready, report compactly:
- Distinctive: line 202
Do not start implementation merely because the specification is complete.
≈ “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
Spec Mode
Turn settled decisions into a buildable specification when no native change workflow already owns the job.
OpenSpec projects delegate early
If the target project already uses OpenSpec and this is a non-trivial code change, load openspec-workflow and delegate immediately.
Do not require all product and architecture decisions to be settled before involving OpenSpec. Native OpenSpec Explore exists specifically to inspect the codebase, weigh options, and clarify change-specific uncertainty before Propose.
In an OpenSpec project, Spec Mode is therefore a compatibility entry point, not a parallel specification system.
Spec Mode is not another planning interview. The default behavior is autonomous: inspect the project, resolve factual gaps, write or refine the specification, verify coherence, and continue until the spec is ready for implementation.
Entry condition
For standalone specification work, start when the important product and architecture decisions are already settled.
For OpenSpec projects, use the delegation rule above instead of this entry condition.
Recover the latest Plan Mode handoff and the target project's current state before doing new work.
If a material decision is genuinely unresolved, do not invent it. Record the exact decision needed and hand it back to Plan Mode.
Minor factual gaps are not planning questions. Retrieve the facts yourself.
Prefer the project's real specification system
If the target project already uses OpenSpec, this mode delegates to openspec-workflow. That workflow inspects the project's real OpenSpec root, schema, existing changes, generated native skills, and available CLI behavior before deciding Explore, Propose, Update, or continuation.
Do not recreate OpenSpec semantics inside Spec Mode.
Default when no schema exists
If the project does not already define a specification system, use the default OpenSpec-style shape:
proposal
↓
specs
↓
design
↓
tasksConceptually:
- Proposal — why the change exists, scope, non-goals, impact;
- Specs — observable requirements and concrete scenarios;
- Design — architecture, interfaces, data flow, constraints, tradeoffs;
- Tasks — implementation slices derived from the settled design.
Keep acceptance evidence explicit.
For behavioral requirements, prefer concrete scenarios such as:
WHEN <condition or action>
THEN <observable outcome>Do not turn implementation guesses into requirements.
Work autonomously
Once the input decisions are sufficient:
- inspect relevant repository files before specifying their replacement;
- research current APIs, libraries, constraints, and external facts when needed;
- reconcile the spec against the actual codebase;
- identify migrations, compatibility concerns, tests, rollout constraints, and failure states;
- eliminate contradictions between artifacts;
- keep scope aligned with the Plan Mode handoff.
Do not repeatedly ask the user to approve routine specification details that are derivable from settled decisions and project facts.
Preserve the decision boundary
Use this rule:
retrievable fact → retrieve it
implementation detail inside settled behavior → specify it
material product/scope decision → return to Plan ModeIf new evidence changes the implementation route without changing intent, update the design coherently.
If new evidence changes the intended behavior, scope, or user-visible contract, stop that branch and request a planning decision instead of silently rewriting intent.
Visual anchor requirement (conditional)
For visually led apps, games, and interfaces, load visual-prototype. Use the approved cohesive concept board, user reference, or current design source of truth as the visual anchor before finalizing the buildable design. If visual exploration must follow an initial exploratory spec draft, incorporate the accepted reference before implementation begins. Keep the exact asset/reference plus visual contract and visual acceptance checks in the canonical project specification. Do not treat extra unapproved images, invented image features, or a text-only prompt as replacements. Stop generating when the first strong visual is accepted. Nonvisual changes bypass this requirement; do not create artwork without a user-established image task.
Definition of ready
The specification is implementation-ready when:
- scope and non-goals are explicit;
- requirements are observable and non-contradictory;
- important edge cases and failure behavior are covered;
- architecture and interfaces are clear enough to implement;
- tasks are derived from the design rather than invented independently;
- acceptance criteria are testable;
- repository facts used by the spec match the current project;
- unresolved product decisions are empty or explicitly deferred;
- Dev Mode can start without reconstructing intent from chat history.
Before declaring completion, perform a coherence pass across all artifacts.
Durable state
Spec Mode may span multiple WebUI turns.
Keep the canonical spec in the target project's existing source of truth whenever writes are authorized.
A continuation checkpoint must preserve:
- target project/repository;
- specification/change identifier;
- artifacts completed;
- artifacts still incomplete;
- evidence/research already gathered;
- unresolved blockers;
- exact next operation.
Do not create a parallel shadow spec merely to checkpoint progress.
Tool-budget continuity
ChatGPT WebUI may hard-stop a tool-heavy assistant turn.
Observed on 2026-09-18: a GitHub-connector-heavy turn stopped at exactly 64 connector calls before a normal assistant final reply.
Treat 64 as an observed GitHub connector boundary, not a universal guarantee.
For GitHub-heavy Spec Mode turns:
- 1–48 connector calls: normal;
- 49–54: keep durable state current;
- around 56: preferred voluntary checkpoint;
- 57–60: finish only useful atomic work or secure state;
- 61–63: close out and checkpoint;
- 64: do not intentionally reach it.
For other tools or future WebUI behavior, adapt conservatively.
At the voluntary stop:
- save the canonical work first;
- record the exact unfinished operation;
- end with:
╭─ TOOL BUDGET ──────────────╮
│ 🟡 checkpoint saved │
│ → send: continue │
╰─────────────────────────────╯When the user sends continue, resume the exact unfinished operation. Do not restart repository discovery, reread everything indiscriminately, or regenerate completed artifacts.
Fresh-session recovery
When invoked as:
continue spec on <repo>first reload this skill.
Then recover:
- the target repository and current branch/change state;
- the existing spec/OpenSpec artifacts;
- the latest durable checkpoint or Plan Mode handoff;
- completed versus unfinished specification work;
- the exact next operation.
Resume from there.
Do not ask the user to restate information that exists in the repository or durable artifacts.
Completion
When ready, report compactly:
- specification/change identifier;
- canonical artifact locations;
- important deferred items, if any;
- evidence that the spec is coherent;
- the correct next stage: Dev Mode for standalone specs, or the OpenSpec workflow's native implementation step for OpenSpec changes.
Do not start implementation merely because the specification is complete.
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.

