deliver-spec
Skill4 starsChanged 16 days agoMITView on GitHub
Deliver a prepared specification through its tickets, shared repository branches, and final human review. Resume its delivery or address final-review feedback.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- holgerleichsenring/specification-first-agentic-developmenttheir own version, first seen Oct 28, 2026
- chrisbanes/skillstheir own version, first seen Jun 18, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- chrisbanes/skills★ 1.1ksame name
- holgerleichsenring/specification-first-agentic-development★ 41same 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 repository31 other files it gives its agents.
- AGENTS.mdAGENTS.md
- ask-smittSkill
- change-walkthroughSkill
- closeoutSkill
- codebase-designSkill
- code-reviewSkill
- deliver-ticketSkill
- diagnosing-bugsSkill
- domain-modelingSkill
- grill-with-docsSkill
- implementSkill
- improve-codebase-architectureSkill
- prototypeSkill
- prSkill
- researchSkill
- resolving-merge-conflictsSkill
- review-feedbackSkill
- setup-federated-workflowSkill
- tddSkill
- to-specSkill
- to-ticketsSkill
- triageSkill
- wayfinderSkill
- wizardSkill
- grillingSkill
- grill-meSkill
- handoffSkill
- teachSkill
- to-questionnaireSkill
- wait-whatSkill
- writing-for-agentsSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 9
Use deliver-ticket for each ticket, closeout for completed tickets, Standalone implement for specification corrections, and code-review for …
- Distinctive: line 11
Use repository-owned bindings for provider operations.
- Distinctive: line 51
After confirmed tracker closure, reconsider dependents without waiting for final specification merge.
- Distinctive: line 53
Complete when: every agreed ticket has confirmed completion, or remaining tickets have attributable active executions, dependency waits, or …
- Distinctive: line 67
Complete when: all specification criteria and obligations have current passing evidence for one fixed combined result, and independent …
- Distinctive: line 75
Keep the Specification open and retain its branches and worktrees for feedback and recovery.
- Distinctive: line 77
Complete when: every changed repository has a verified final proposal or preserved confirmed merge, the durable handoff matches current …
≈ “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
Coordinate one published Specification and its prepared delivery tickets. Accept its permanent reference in its Ticket Origin Repository. This invocation authorizes specification branches and proposals, ticket execution with automatic merge into those branches, and ticket closeout when each ticket's own completion criteria pass. Final specification merge and closeout remain human-controlled.
Use deliver-ticket for each ticket, closeout for completed tickets, Standalone implement for specification corrections, and code-review for independent review. These skills must be available as invokable skills. Their bundled instructions are reference material, not installed entry points.
Use repository-owned bindings for provider operations. Preserve host protections, unrelated work, and intermediate human decisions. Keep specification progress on the Specification and detailed ticket execution in its existing checkpoints.
1. Establish authority and composition
Read the Ticket Origin Repository's instructions and applicable Work Tracker and routing bindings. On an existing execution, use recovery to determine the next action before acquiring context. For specification scheduling or validation, follow Context Pack consumption to obtain or reuse Complete Decision Context for the exact Specification reference. Read the Specification and applicable accepted changes; retain its entry path for consumers. Publication owns planning validation. A discovered gap or changed authority requires reconciliation before affected work continues.
Discover the complete explicitly related delivery-ticket set through the configured relationship procedure, including every page. Decision Context supplies authority, not a complete scheduler snapshot. For Local Markdown, enumerate a declared ticket index or explicitly supplied relationship directory when available. Otherwise enumerate the established .scratch/*/issues/*.md layout in the Ticket Origin Repository. Read ticket relationship fields and select exact permanent Parent or Authority Sources references to this Specification; directory proximity alone establishes no relationship. Resolve repository-relative references and enumerate the chosen set completely. This uses ticket-side relationships and requires no reverse index on the Specification. Missing relationships or an unavailable provider procedure require clarification, not guessed tickets.
Read each related ticket's published relationship fields, Repository References and writable Repository Scope, plus current dependency, claim, routing, and completion state. Use these records for the complete scheduling set and writable repository union. This is scheduling discovery, not per-ticket Context Assembly or a planning audit. Acquire each ticket's Delivery Context in its deliver-ticket execution when selected for launch. A Specification's Repository References alone grant no writes. Missing or contradictory scheduling records stop affected scheduling and return to their owner; this skill does not create substitute tickets.
Record the agreed permanent ticket set, effective authority references, writable repository union, and supplied execution directions in a specification checkpoint. On any existing execution, apply recovery before mutation. New tickets or changed requirements require explicit reconciliation. Pause affected work while valid independent work may continue. Preserve completed history; record approved changes and their authority before rebuilding affected packs.
Complete when: accepted authority and the exhaustive agreed scheduling set are available, every writable repository has attributable authorization, and existing execution or changed composition has been reconciled.
2. Prepare shared branches and execution contexts
Prepare shared targets for eligible tickets as capacity becomes available. Resolve the needed repositories against their Repository IDs, declared remotes, selected final targets, Git identities, and Code Host bindings. Read instructions needed for branch preparation. Domain implementation reading belongs to ticket executors; final validation preparation belongs to step 4. Use a matching known checkout as a Git store while preserving its checked-out branch, index, files, and configuration. Unresolved identity or access stops affected preparation without blocking independent tickets whose prerequisites are ready.
Record one specification integration branch and final Target Branch per writable repository, normally using the configured Base Branch as final target. For each repository needed by the next launch, pin the final target's current exact commit. Create the integration branch from that commit only after verifying the intended name is unused and the target exists. Publish through the binding and verify the remote revision before dispatching its ticket. A colliding branch requires ownership proof or reconciliation, never overwrite. Persist branch choices before creation and verified revisions afterward; future repository preparation remains explicitly pending.
Reuse attributable branches and suitable isolated specification worktrees. Create a worktree only when validation or corrections need writable local state; obey the execution host's location policy and repository bootstrap instructions. Record ownership separately from ticket worktrees. Durable records use repository, branch, revisions, and provenance; local paths remain host state. Keep read-only repositories outside the writable union and isolate mutable validation outputs from their stores.
Reconcile adopted results and active deliveries through recovery. A completed ticket must be represented in the assembled result. An active delivery on another target needs explicit reconciliation before reuse; never silently retarget it.
Select a context mechanism that can start, observe, continue, and recover ticket coordinators which themselves delegate implementation and review. Follow user preferences and actual runtime capabilities. When the user selects separate chats in the same project, use those chats for ticket coordinators and their subagents for repository work. Verify that capability before dispatch. If unavailable, report the limitation and obtain an authorized alternative rather than flattening context isolation.
Budget capacity across the delegation tree. Independent tickets may run in parallel, including two tickets writing the same repository through separate ticket worktrees. Shared repository identity is not a scheduling dependency. Retain execution references for reuse; runtime preferences belong to the user and execution host, not repository setup.
Complete when: the next eligible ticket's shared targets are durable and remotely readable, its adopted prerequisites are accounted for, and the context mechanism supports the required isolation and recovery within actual capacity. Dispatch it while preparing other independent eligible work.
3. Execute and close eligible tickets
Use current tracker dependency rules to select only eligible tickets in the agreed set. Recheck state before dispatch. Record launch intent and its ticket before starting a coordinator; record and re-read the returned execution reference afterward. Use the runtime's available lookup to reconcile an uncertain launch before any retry.
Give each coordinator an explicit ticket reference, Ticket Origin Repository, and Automatic merge: allowed. Supply Target Branch entries only for repositories in that ticket's published Repository Scope, using each repository's recorded specification branch. Include relevant orchestration decisions and resource ownership. An already accepted matching ticket pack may be passed explicitly with its complete directory; do not acquire one merely to dispatch. Invoke deliver-ticket in that separate context. It owns live eligibility and claim verification, its Delivery Context, target synchronization, conflicts, validation, review, merges, handoff, and ticket recovery. Its isolated ticket source branches and per-ticket proposals remain part of delivery. Scheduling state does not replace its pre-claim check.
Observe or continue existing executions through their recorded references. For a ticket human-action checkpoint, retain the ticket's request as the response location and link it from the Specification. Surface the required action and pause affected and dependent work. Continue independent work only when its authority and prerequisites remain valid. A response resumes through deliver-ticket's gate. A specification-level question belongs on the Specification with the affected work, required decision, owner, response location, and completion condition.
After a successful ticket handoff with confirmed merges or valid no-change results, check that ticket's actual completion criteria. Invoke closeout with exact revision evidence only when they pass. Explicitly scope cleanup to that ticket's disposable resources and identify specification branches and worktrees as retained shared resources. Merge alone supplies neither missing acceptance nor required deployment evidence. Preserve the human boundary until those conditions are met.
After confirmed tracker closure, reconsider dependents without waiting for final specification merge. Distinguish closure from cleanup; incomplete cleanup need not undo proven closure. Recover missing effects through their existing owner and update the specification checkpoint.
Complete when: every agreed ticket has confirmed completion, or remaining tickets have attributable active executions, dependency waits, or human boundaries. Continue eligible work while capacity permits; otherwise return the current checkpoint and required next actions.
4. Validate the assembled result
Once all agreed tickets are complete, pin the complete repository head set, current final target commits and review bounds, and every influencing Validation Source revision. Verify the assembled result includes all adopted ticket outcomes. Freeze coordinator writes while validating; observed external movement follows evidence freshness.
For each unmerged changed repository, synchronize the current final target into the isolated specification branch if it advanced. Preserve concurrent work. Resolve only conflicts determined by accepted requirements; a semantic conflict requiring new authority enters the specification human boundary. Record the new review base and resulting head before validation.
Run repository validation and all specification acceptance criteria and declared Validation Obligations against that exact combined result. Assign each obligation one evidence owner and its declared method, prerequisites, expected result, and exact influencing revisions. Reuse ticket evidence only when those complete dependencies still match. A clean set of ticket checks does not prove the combined result. Planned human verification remains a visible requirement with its declared procedure and evidence.
Use code-review in Both mode on fixed Repository Targets for changed repositories. Supply the accepted Specification Decision Context unchanged, exact review bases and heads, settled seams, repository evidence, and whole-specification validation. Include unchanged and confirmed merged repositories as exact read-only authority or Validation Sources where they influence acceptance. An empty diff requires no empty review target or proposal, but still requires the applicable acceptance evidence.
Record Advisory findings. When an attributable combined-result failure or Blocking review finding requires an in-scope correction, read and follow corrections, then refresh affected validation and independent reviews. Requirement or scope changes require the human boundary.
Complete when: all specification criteria and obligations have current passing evidence for one fixed combined result, and independent review has no Blocking finding. Otherwise preserve the result with its exact failed or pending conditions.
5. Publish the final human handoff
For each unmerged changed repository, publish the specification branch through its binding and create or update its one final Review Proposal to the recorded final target. Reconcile existing proposals before creation. Re-read source, target, remote head, draft/readiness state, feedback, and host checks. Any drift returns to the earliest invalidated prerequisite before claiming readiness. Mark verified proposals ready for human review. Never merge them automatically or bypass host policy.
Preserve confirmed merged proposals with their exact source/result revisions. Repositories with no resulting changes need no proposal. Record a current handoff checkpoint containing final proposals, validation and independent-review results, limitations, and remaining human review, merge, rollout, acceptance, and explicit closeout actions. Reuse an equivalent verified handoff on repeat invocation.
Keep the Specification open and retain its branches and worktrees for feedback and recovery. Do not invoke specification closeout. When actionable final-review feedback or a later correction request arrives, read and follow recovery, then corrections, for this same Specification.
Complete when: every changed repository has a verified final proposal or preserved confirmed merge, the durable handoff matches current evidence, and the human has the remaining actions with recoverable working state.
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.

