spec-driven
Skill4 starsChanged 3 months agoMITView on GitHub
Apply or explain the lifecycle after a deliverable specification is greenlit: Lead Agent coordination, Contributor Agent implementation, Deliverable Owner steering, AC/TDD, in-place spec contract updates, run logs, QA/UAT, PR feedback, re-verification, and final human sign-off. Use for questions about how the specification-driven cycle coordinates Project Owners, Deliverable Owners, Contributors, and AI agents.
14 other projects use a version of this.
Adopted by
Most recently taken up by
- adeonir/agent-skillstheir own version, first seen Nov 6, 2026
- Snowflake-Labs/coco-skillstheir own version, first seen Nov 1, 2026
- NoatToan/claude-code-pluginstheir own version, first seen Nov 1, 2026
- stoa-platform/stoatheir own version, first seen Nov 1, 2026
- microsoft/FSQtheir own version, first seen Oct 13, 2026
- aprameyak/coding-skillstheir own version, first seen Oct 2, 2026
Used with
Projects with it also use implement.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
Hugging FaceSkillimplement
Make tests pass. Invoke after /write-tests produces failing tests.
Also in 3 projects that use this
Other versions14 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 14
- agencia-conversion/agentic-seo-skills★ 51same name
- Snowflake-Labs/coco-skills★ 49same name
- microsoft/FSQ★ 23same name
- stoa-platform/stoa★ 9same name
- adeonir/agent-skills★ 8same name
- creoaurea-devlab/agentic-skills★ 4same name
- 90sRehem/agent-skills★ 2same name
- b12consulting/skills★ 2same name
- hbarbosa02/alaz-mcp-nestjs★ 2same name
- aprameyak/coding-skills★ 1same name
4 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 repository21 other files it gives its agents.
- templates/AGENTS.mdAGENTS.md
- architecture-doc-builderSkill
- confluenceSkill
- conformanceSkill
- doc-checkSkill
- feature-doc-builderSkill
- gh-issueSkill
- gh-stackSkill
- gh-wikiSkill
- handoffSkill
- jiraSkill
- pr-builderSkill
- proposal-builderSkill
- pr-reviewSkill
- psqlSkill
- qa-testingSkill
- spec-builderSkill
- sqlcmdSkill
- standard-builderSkill
- worktreeSkill
- zazz-boardSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
26 rules. 0 are common, 18 are distinctive, and 8 are very specific to this project.
0
Common
Most of the 12 edited versions we read keep these.
18
Distinctive
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 8
Use this skill for the lifecycle around an approved deliverable specification from greenlight through human-reviewed merge.
- Very specific: line 14
If the user is drafting or refining the initial specification before greenlight, use spec-builder.
- Distinctive: line 67
Treat steps as a controlled feedback loop.
- Distinctive: line 75
read the greenlit specification and required references
- Distinctive: line 76
verify applicable standards from the repo standards index
- Distinctive: line 78
start with the specified TDD entry point or strongest narrow verification
- Distinctive: line 80
run the named tests or manual checks
- Distinctive: line 84
If an implementation detail can adapt without changing scope, public behavior, ACs, test strategy, approved review shape, or invariants, …
- Distinctive: line 102
run-log and tracker updates
- Distinctive: line 116
Prefer fresh-context agents for independent QA and verification passes.
- Distinctive: line 129
Other tracker: follow the repo-declared tracker workflow.
- Distinctive: line 130
Local run log only: keep the run log and PR evidence current.
- Very specific: line 134
Use this protocol when accepted feedback changes scope, public behavior, UX workflow, API/schema/validation behavior, bug-fix behavior, …
- Distinctive: line 138
Stop coding the affected slice.
- Distinctive: line 141
Update the affected specification sections in place so they describe the current contract.
- Distinctive: line 150
Use this entry shape:
- Distinctive: line 182
test command output with no contract change
- Distinctive: line 193
Create or request a separate specification when accepted feedback:
- Distinctive: line 205
Before marking implementation ready for human review:
- Distinctive: line 207
every current AC has evidence
- Very specific: line 219
Use spec-builder for initial specification authoring, templates, intake, topology, ACs, test strategy, and implementation prompt shape.
- Distinctive: line 221
Use this skill for post-greenlight lifecycle, contract-change handling, evidence flow, and sign-off discipline.
- Very specific: line 223
Use qa-testing for independent verification and evidence quality.
- Very specific: line 224
Use pr-builder for draft-first PR packaging.
- Very specific: line 225
Use pr-review for standards/spec compliance review.
- Very specific: line 226
Use gh-stack for stacked branch command details when the approved review shape is a stack.
“Common” is measured: kept by at least 6 in 10 of the 12 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
Spec-Driven Development
Use this skill for the lifecycle around an approved deliverable specification from
greenlight through human-reviewed merge. It does not replace spec-builder,
qa-testing, pr-builder, pr-review, gh-stack, or Contributor Agents. It defines
how those roles coordinate around the specification as the current implementation
contract.
If the user is drafting or refining the initial specification before greenlight, use
spec-builder. If the user is asking how an approved specification moves through
implementation, steering, QA, PR feedback, controlled contract updates, rework,
re-verification, and sign-off, use this skill.
Core Model
A deliverable specification is the executable contract for one deliverable. After greenlight, the body of the specification should continue to read as the current contract. Accepted Deliverable Owner steering, QA/UAT findings, PR review feedback, or implementation-discovered bugs that change the contract update the affected specification sections in place.
The final Implementation And Review Change Log records the audit trail for those
changes. It points to changed sections and explains source, rationale, summary, and
verification impact. It is not a competing list of current requirements.
Append-only execution history still belongs in the run log or repo-declared execution record. The run log grows as work proceeds; do not rewrite prior entries unless the user explicitly asks. The run log records attempts, phase progress, failures, evidence, QA findings, handoffs, and recoveries. The specification records the current contract plus the final contract-change audit trail.
Roles In This Cycle
- The Project Owner remains accountable for project direction and durable product context.
- The Deliverable Owner is the human authority for a deliverable's scope, contract changes, acceptance, and review or merge readiness.
- A Contributor is a human engineer contributing to the deliverable.
- A Contributor Agent is an AI agent explicitly delegated bounded work.
- The Lead Agent is a designated Contributor Agent. It coordinates execution, delegates to other Contributor Agents and Verifier Agents, integrates their output, and maintains execution evidence. It does not assume Deliverable Owner authority.
- A Verifier Agent or QA Agent independently evaluates work and does not modify implementation unless explicitly reassigned.
Lifecycle
flowchart TD
A["Spec creation and refinement<br/>(spec-builder)"] --> B["Deliverable Owner greenlights specification"]
B --> C["Implement through AC/TDD loop"]
C --> D{"Deliverable Owner steering,<br/>QA/UAT, PR review,<br/>or discovered bug changes contract?"}
D -- "No" --> E["Record progress and evidence<br/>in run log"]
E --> F["Draft PR and author-side review"]
F --> G{"Review feedback changes contract?"}
G -- "No" --> H["Rework implementation only<br/>and re-verify"]
H --> I["Final evidence and human sign-off"]
I --> J["Merge through human PR process"]
D -- "Yes" --> K["Get Deliverable Owner sign-off"]
G -- "Yes" --> K
K --> L["Update affected spec sections in place"]
L --> M["Append Implementation And Review Change Log entry"]
M --> N["Update run log with execution details"]
N --> CTreat steps as a controlled feedback loop. Implementation, review, and rework may cycle several times, but every accepted contract change leaves both the current spec body and the audit trail coherent.
AC/TDD Implementation Loop
The Lead Agent carries out or delegates this loop to Contributor Agents:
- read the greenlit specification and required references
- verify applicable standards from the repo standards index
- resolve open questions before editing
- start with the specified TDD entry point or strongest narrow verification
- implement in small slices against acceptance criteria
- run the named tests or manual checks
- record phase progress, evidence, deviations, and recoveries in the run log
- repeat until each current AC has evidence
If an implementation detail can adapt without changing scope, public behavior, ACs, test strategy, approved review shape, or invariants, log meaningful deviations in the run log and continue. If the contract changes, use the change protocol below.
Lead Agent, Contributor Agents, And Tracking
The implementation prompt should name the coordination model. The Lead Agent may work alone or delegate bounded work to Contributor Agents when the active harness supports delegation and the specification allows it. A harness may call delegated Contributor Agents subagents, but Contributor Agent is the methodology term.
The Lead Agent is responsible for:
- keeping execution within the current specification
- task/phase delegation boundaries
- file-conflict serialization inside the active worktree
- integration of Contributor Agent output
- evidence quality and AC mapping
- run-log and tracker updates
- final PR-ready output
Contributor Agents may perform bounded phases, tasks, branch slices, tests, QA checks, or documentation only when the specification names those boundaries. They return changed-file summaries, commands run, evidence, risks, and unresolved questions. If they are not available, the Lead Agent performs the phases directly and records that in the run log.
Work runs in a single active worktree or stack lane for the approved review artifact. The Lead Agent orders delegated work so overlapping file work is serialized. Do not let Contributor Agents overwrite one another's edits. When two tasks may touch the same file, sequence the tasks and reconcile the diff before continuing.
Prefer fresh-context agents for independent QA and verification passes. A QA agent may focus on functionality, performance, code hygiene, security, accessibility, standards, or stack no-drift, depending on the specification. QA agents read the current specification, run log/tracker record, relevant evidence, and focused code scope; they do not modify code.
Execution tracking follows the repo's declared system:
- Zazz Board: load
zazz-boardand update implementation progress, delegated Contributor Agent task progress, notes, locks when required, statuses, and evidence links through the board. - Jira: load
jiraand use repo-provided or Deliverable Owner-provided Jira issue context; do not assume live Jira access unless the repo declares it. - Other tracker: follow the repo-declared tracker workflow.
- Local run log only: keep the run log and PR evidence current.
Change Protocol
Use this protocol when accepted feedback changes scope, public behavior, UX workflow, API/schema/validation behavior, bug-fix behavior, test strategy, execution sequence, approved review shape, branch contract, invariants, or acceptance criteria.
- Stop coding the affected slice.
- Identify the sections that need to change.
- Get Deliverable Owner sign-off for the contract change.
- Update the affected specification sections in place so they describe the current contract.
- Append an
Implementation And Review Change Logentry at the end of the specification. - Record execution details in the run log, including rework and re-verification notes.
- Re-run any tests or manual checks whose evidence may have been invalidated.
- For stacked branches, rebase/propagate the updated lower-branch contract upstack and re-run no-drift checks.
Use this entry shape:
### YYYY-MM-DD HH:MM TZ — Short Change Title
**Source.** Deliverable Owner steering, QA/UAT, PR review, implementation-discovered bug, or other source.
**Changed Sections.** Links or section numbers.
**Rationale.** Why the accepted change was needed.
**Summary.** Short summary of the in-place spec edits.
**Verification Impact.** Tests, manual checks, evidence, or re-verification now required.What Belongs Where
- Specification body: current executable contract. Update it in place when accepted feedback changes the contract.
- Implementation And Review Change Log: final spec section; audit trail for accepted contract changes after greenlight.
- Run log / execution record: append-only execution history, attempts, phase progress, evidence locations, failures, recoveries, QA findings, handoffs, and rework notes.
- PR body / review thread: reviewer-facing summary, evidence, open risks, and review conversation. Summarize material spec changes and link the specification.
- Commit history: implementation history. Do not rely on commits alone to explain contract changes.
Do Not Use The Change Log For
- ordinary progress notes
- test command output with no contract change
- pure refactors that preserve contract
- formatting, naming, or README cleanup unrelated to the deliverable's product contract
- PR-body-only edits
- temporary failed attempts
- implementation details that remain within adaptive guidance
Those belong in the run log, PR, or commit history as appropriate.
When To Create A New Spec Instead
Create or request a separate specification when accepted feedback:
- creates a new deliverable boundary
- needs a separate review artifact
- changes branch/PR topology beyond the approved review shape
- introduces a new feature context
- makes the original deliverable misleading even after in-place updates and audit log
- is primarily docs, process, tooling, or standards work outside the deliverable's product-facing capability
Review And Sign-Off
Before marking implementation ready for human review:
- every current AC has evidence
- affected spec sections match the implemented behavior
- each contract change has a final change-log entry
- the run log is current through rework and re-verification
- QA or verifier findings are resolved or explicitly carried as known risk
- draft PR evidence links to the specification and summarizes material contract changes
Humans retain scope approval, subjective product/UX sign-off, PR approval, and merge authority.
Skill Boundaries
- Use
spec-builderfor initial specification authoring, templates, intake, topology, ACs, test strategy, and implementation prompt shape. - Use this skill for post-greenlight lifecycle, contract-change handling, evidence flow, and sign-off discipline.
- Use
qa-testingfor independent verification and evidence quality. - Use
pr-builderfor draft-first PR packaging. - Use
pr-reviewfor standards/spec compliance review. - Use
gh-stackfor stacked branch command details when the approved review shape is a stack. A gh-stack lane keeps every stacked branch in one dedicated worktree; each branch and PR represents one dependent deliverable.
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.

