task-decomposition
Skill4 starsChanged 29 days agoMITView on GitHub
[METHODOLOGY] Approved tech-spec → atomic engagement tasks. SDD phase 3: dev-lead applies it in the `{name}-impl` engagement when it maps the tech-spec's Implementation Tasks onto the plan. Preloaded by dev-lead and dev-tech-architect.
18 other projects use a version of this.
Adopted by
Most recently taken up by
- MMACABRE/SKILLStheir own version, first seen Nov 8, 2026
- michaelalber/ai-toolkittheir own version, first seen Nov 1, 2026
- mabzhang/opc-toolkittheir own version, first seen Nov 1, 2026
- DevelopersGlobal/ai-agent-skillstheir own version, first seen Oct 25, 2026
- formal-you/agent-skills-suitetheir own version, first seen Oct 20, 2026
- Mushfiqur-Rahman-Robin/coding-rulestheir 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
- DevelopersGlobal/ai-agent-skills★ 216same name
- adamentwistle/fable-skills★ 98same name
- charliehzm/medharness★ 85same name
- baekenough/oh-my-customcode★ 42same name
- d-oit/rust-2026-template★ 15same name
- aws-solutions/konductor★ 11same name
- hieutrtr/ai1-skills★ 8same name
- karthikrshet/aiskills★ 4same name
- Lukasbrujula/claude-code-framework★ 4same name
- mabzhang/opc-toolkit★ 4same 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 repository48 other files it gives its agents.
- skills/dev-methodology/references/agents.mdAGENTS.md
- skills/skill-authoring/references/agents.mdAGENTS.md
- acceptance-protocolSkill
- agency-intakeSkill
- ai-visibility-methodologySkill
- banner-design-guideSkill
- benchmark-researchSkill
- brand-methodologySkill
- code-reviewingSkill
- code-writingSkill
- codex-bridgeSkill
- deploy-pipelineSkill
- design-assets-guideSkill
- design-system-methodologySkill
- design-task-decompositionSkill
- dev-methodologySkill
- docs-pipelineSkill
- documentation-writingSkill
- engagement-contractSkill
- engagement-protocolSkill
- feature-researchSkill
- human-voiceSkill
- infrastructure-setupSkill
- marketing-task-decompositionSkill
- post-deploy-qaSkill
- pre-deploy-qaSkill
- presentation-designSkill
- project-planningSkill
- prompt-engineeringSkill
- security-auditingSkill
- semantic-drift-methodologySkill
- seo-auditingSkill
- skill-authoringSkill
- skill-test-designSkill
- skill-testingSkill
- system-optimization-protocolSkill
- tech-spec-planningSkill
- testing-methodologySkill
- ui-styling-guideSkill
- ui-ux-methodologySkill
- user-spec-planningSkill
- validation-pipelineSkill
- yandex-analytics-methodologySkill
- yandex-direct-guideSkill
- yandex-metrika-guideSkill
- yandex-search-guideSkill
- yandex-webmaster-guideSkill
- yandex-wordstat-guideSkill
The file
We split this skill into the rules it gives an agent
6 rules. 0 are common, 2 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 14 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 41
Read engagement/specs/tech-spec.md.
- Very specific: line 60
One to one, unless a recorded drift says otherwise: Do not merge, split or drop tech-spec tasks on your own judgement.
- Very specific: line 83
Always add completeness-validator (traceability user-spec ↔ tech-spec ↔ code) and anti-pattern-detector.
- Distinctive: line 88
Before returning the plan:
- Very specific: line 89
Coverage (§6 of the integrity reference): Every AC-N and every TC-N reaches at least one plan task through crit_refs, or, for a TC the …
- Distinctive: line 93
Disjoint files per wave: No two same-wave tasks share a file.
“Common” is measured: kept by at least 6 in 10 of the 14 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
Task Decomposition
Turn the approved tech-spec's Implementation Tasks into the {name}-impl engagement's plan tasks
without losing intent.
Input: engagement/specs/tech-spec.md (status: approved) + engagement/specs/user-spec.md
Output: plan.tasks / plan.waves returned by dev-lead; the engine's decompose phase writes
engagement/tasks/*.md and tasks/INDEX.md from them.
Where this runs. Inside engagement-workflow, at the lead:plan step of the impl engagement.
Nobody here can reach the user, and nobody spawns subagents: the older multi-agent flow
(task-creator fan-out, task-validator + reality-checker batches, user approval of the task
set) belonged to the retired work/{feature}/ pipeline and does not run inside an engagement.
The checks it performed now happen at three engine seams: your self-check below, the per-task
review (and the co-signed contract when A.contracts is on), and the validate phase.
Integrity rules (read first)
Decomposition is a lossy migration of intent, not a summary. The rules that keep meaning intact across the conversion live in one shared file: decomposition-integrity.md.
Load it before mapping. Non-negotiable subset: every task names its Source and repeats
the Common objective; the source wins on any conflict with task wording; requirements that
exist because of a risk carry that risk into ## Non-goals / must not break; small same-shape
edits batch into ONE task instead of one seat each; the coverage check in §6 runs before the set
is handed over.
Step 1: Gate on approval
Read engagement/specs/tech-spec.md. Frontmatter must say status: approved with a non-null
approved_by. If it does not, return a plan with zero delivery tasks and put
SDD gate: tech-spec not approved in notes: the impl engagement does not start on a draft.
Step 2: Map tasks one to one
For each Implementation Task in the code waves (everything before Audit Wave and Final Wave):
| tech-spec field | plan task field |
Task N: {Name} | id = T{N}, title = {Name} [tech-spec Task N] |
| Owner | owner |
| Files to modify | files (repo-relative, unchanged) |
Covers (AC-N, TC-N) | crit_refs = the crit-N ids of the impl criteria.md Done-when bullets that cite those AC/TC ids (bullets read - crit-3 (AC-2): …) |
| Depends on | depends_on |
| Wave | the task's position in plan.waves |
| Description, Verify-smoke, Verify-user, Files to read | not copied: the plan schema has no per-task field for them. The [tech-spec Task N] marker in the title and the impl criteria.md Scope line send the specialist and the reviewer to that task in specs/tech-spec.md |
Rules:
- One to one, unless a recorded drift says otherwise. Do not merge, split or drop tech-spec
tasks on your own judgement. The only deviations are the cases below, and each one is written
as
SDD driftso the manager adjudicates it. A task that cannot run as written (overlapping files in one wave, a missing owner, a file that does not exist) is a spec defect: record it innotesandscope-sync.mdasSDD drift: {task} — {what}and plan the nearest executable shape. The manager adjudicates drift at acceptance. A path marked(new)is expected not to exist yet; drop the marker infiles. - Batching (drift). §5 of the integrity reference still applies: if the tech-spec gave several same-shape one-line edits a seat each, batch them and record the batch as drift.
- No new work (drift). A task that covers no AC/TC and is not infrastructure another task depends on does not enter the plan; record the dropped task as drift.
Step 3: Map Audit Wave and Final Wave onto the validate phase
| tech-spec declaration | engagement plan |
| Code Audit | code-reviewer in plan.validators |
| Security Audit | security-auditor in plan.validators |
| Test Audit | test-reviewer in plan.validators |
| QA (pre-deploy) | pre-deploy-qa in plan.validators |
| Deploy | a dev-devops-engineer task in the last delivery wave |
| Post-deploy verification | post-deploy-qa in plan.validators |
Always add completeness-validator (traceability user-spec ↔ tech-spec ↔ code) and
anti-pattern-detector.
Step 4: Coverage and cross-task self-check
Before returning the plan:
- Coverage (§6 of the integrity reference). Every
AC-Nand everyTC-Nreaches at least one plan task throughcrit_refs, or, for a TC the tech-spec assigns to an Audit/Final declaration, the matching validator inplan.validators. An uncovered id is a hole, not a judgement call. - Disjoint files per wave. No two same-wave tasks share a file.
- Shared resources. Each heavy resource in tech-spec Architecture → Shared resources has one owning task; its consumers depend on that task.
Per-task contract (S2 — when A.contracts is enabled)
On M/L engagements run with A.contracts on, each tasks/{id}.md ALSO carries a ## Contract (co-signed) section (the per-task acceptance rubric). You do NOT author it — leave it for the engine's contract handshake (owner proposes assertions in-band → reviewer co-signs and writes the section). Schema: engagement-protocol §"tasks/{id}.md → ## Contract (co-signed)". It binds the per-task review only and never waives criteria.md.
Final Check
- tech-spec approved (
status: approved,approved_byset) before any task was planned - Every code task mapped one to one; every deviation recorded as
SDD drift - Audit/Final waves mapped onto
plan.validators(+ deploy task if declared) - Every
AC-N/TC-Ncovered; same-wave files disjoint
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.

