task-decomposition
Skill1 starsChanged 9 months agoMITView on GitHub
Decompose approved tech-spec into atomic task files with parallel creation and validation.
19 other projects use a version of this, 1 of them edited.
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
- AgentShekel/agentic-workflowtheir own version, first seen Oct 14, 2026
Other versions19 projects use a version of this.
Word for word · 0
None found.
Edited · 1
- stepanenkoviktor0110-boop/ai-dev-methodology★ 232% same
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
- AgentShekel/agentic-workflow★ 4same name
- karthikrshet/aiskills★ 4same name
- Lukasbrujula/claude-code-framework★ 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 repository20 other files it gives its agents.
- skills/skill-master/references/agents.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- code-reviewingSkill
- code-writingSkill
- deploy-pipelineSkill
- documentation-writingSkill
- feature-executionSkill
- infrastructure-setupSkill
- methodologySkill
- post-deploy-qaSkill
- pre-deploy-qaSkill
- project-planningSkill
- prompt-masterSkill
- security-auditorSkill
- skill-masterSkill
- skill-test-designerSkill
- skill-testerSkill
- tech-spec-planningSkill
- test-masterSkill
- user-spec-planningSkill
The file
We split this skill into the rules it gives an agent
11 rules. 0 are common, 6 are distinctive, and 5 are very specific to this project.
0
Common
Most of the 15 edited versions we read keep these.
6
Distinctive
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 20
Ask user for feature name if not provided.
- Very specific: line 22
Read work/{feature}/tech-spec.md.
- Very specific: line 23
If not approved — tell user: "tech-spec не утверждён.
- Distinctive: line 29
Read skills/reviewers catalog from skills-and-reviewers.md — for passing correct skills/reviewers to task-creators.
- Distinctive: line 69
Read JSON reports, collect findings.
- Very specific: line 70
If issues found — for each task with issues, launch task-creator in fix mode:
- Very specific: line 73
After each validation round, git commit: chore(tasks): validation round {N} — {summary}
- Distinctive: line 75
If problems remain after 3rd iteration — show user: "Вот что осталось — давай решим вместе."
- Distinctive: line 79
After individual validation passes, run a final cross-task check:
- Very specific: line 85
If issues found → launch task-creator in fix mode for affected tasks.
- Distinctive: line 96
Wait for user approval.
“Common” is measured: kept by at least 6 in 10 of the 15 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
Decompose tech-spec Implementation Tasks into individual task files with parallel creation and validation.
Input: work/{feature}/tech-spec.md (status: approved)
Output: work/{feature}/tasks/*.md (validated)
Language: Task files in English, communication in Russian
Phase 1: Create Tasks
- Ask user for feature name if not provided.
- Read
work/{feature}/tech-spec.md. Check frontmatterstatus: approved. If not approved — tell user: "tech-spec не утверждён. Сначала запусти/new-tech-specи доведи до approved." Stop. - Read
work/{feature}/user-spec.md. - Note the task template path:
~/.claude/shared/work-templates/tasks/task.md.template - Read skills/reviewers catalog from skills-and-reviewers.md — for passing correct skills/reviewers to task-creators.
- For each task in Implementation Tasks — launch
task-creatorsubagent in parallel. Pass each task-creator:- feature_path, task_number, task_name
- template_path:
~/.claude/shared/work-templates/tasks/task.md.template - files_to_modify, files_to_read (from tech-spec)
- depends_on, wave, skills, reviewers, verify (from tech-spec)
- teammate_name (if specified in tech-spec, optional)
Each task-creator copies the template to
tasks/{N}.mdfirst, then edits each section in place. This ensures no sections are skipped.
- Confirm each task-creator returned a file path. Skip reading task content — preserve context budget for validation phase.
- Git commit:
draft(tasks): create {N} tasks from tech-spec for {feature}
Checkpoint:
- All
tasks/*.mdfiles created - Each task-creator returned file path
- Draft committed
Phase 2: Validation (up to 3 iterations)
Tech-spec was already validated by 6 validators. This phase checks only: (1) task-creator correctly expanded tasks by template, (2) no mismatches with real code appeared during detailing.
Validators
Launch both in parallel:
task-validator (sonnet) — Template Compliance + AC/TDD carry-forward:
- Batch: 5 tasks per call
- Pass: feature_path, task_numbers array, batch_number, iteration
- Report:
logs/tasks/template-batch{N}-review.json
reality-checker (sonnet) — Reality & Adequacy:
- Batch: 3 tasks per call
- Pass: feature_path, task_numbers array, batch_number, iteration
- Report:
logs/tasks/reality-batch{N}-review.json
Process
- Launch both validators in parallel (task-validator in batches of 5, reality-checker in batches of 3).
- Read JSON reports, collect findings.
- If issues found — for each task with issues, launch
task-creatorin fix mode:- Pass: same inputs as creation +
mode: fix+findingsfrom validators - task-creator reads existing task, applies fixes, overwrites file
- Pass: same inputs as creation +
- After each validation round, git commit:
chore(tasks): validation round {N} — {summary} - Re-validate fixed tasks (repeat 1-4). Maximum 3 iterations.
- If problems remain after 3rd iteration — show user: "Вот что осталось — давай решим вместе."
Cross-Task Integration Check
After individual validation passes, run a final cross-task check:
- Launch both validators on ALL tasks in a single batch (not split into smaller batches):
task-validator— focus: shared resource ownership (one owner, consumers depend_on owner), no competing instances in same wavereality-checker— focus: duplicate heavy resource init, hidden dependencies, inconsistent approaches across tasks
- If issues found → launch
task-creatorin fix mode for affected tasks. Re-validate fixed tasks. - Max 2 iterations for cross-task check (on top of the 3 individual iterations).
Checkpoint:
- Both validators: status=approved OR user resolved remaining issues
- Cross-task integration check: no cross-task conflicts
Phase 3: Present to User
- Summary: task count, waves, dependencies, validation results (iterations, issues found/fixed).
- Wait for user approval.
- Git commit:
chore(tasks): task decomposition approved for {feature} - Suggest next step:
/do-taskfor individual tasks.
Checkpoint:
- Summary presented to user
- User approved task decomposition
- Approval committed
Final Check
- All phases completed (tasks created, validation passed)
- All tasks match template (frontmatter: status, depends_on, wave, skills, reviewers, teammate_name)
- Validation: both validators passed or user confirmed remaining issues
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.

