Decompose backlog
Skill theocarranza/agile-backlog-toolkit/skills/decompose-backlog
Decompose a parent work item into correctly-parented, audited child Stories. Use when the user asks to "break down this Feature", "decompose Epic/Feature N into stories", "create the user stories for <feature>", "groom this backlog item", or provides a Feature/Epic id and wants Stories drafted and created in Azure DevOps. Drives 7 phases: ingest the parent, split into right-sized Stories (1 Story = 1 sprint = 1 PR), draft them in the artifacts path, enrich to the team format, create them in Azure DevOps parented to the FEATURE, verify the hierarchy, and audit that every parent requirement has a home. Self-contained rules; two approval gates before any write. For plain-language scope lines and story narrative, delegates sub-passes to generate-plain-language-documentation in DRAFT and ENRICH phases.From its SKILL.md
npx -y skills add theocarranza/agile-backlog-toolkit --skill decompose-backlogAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its file declares
Copied from the file, not written here
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
5.7 KB, as published. Nobody here has run it
Decompose Backlog
Conductor for turning a parent work item into child Stories in Azure DevOps. Load the reference files as each phase needs them — they carry the self-contained rules so this file stays a score, not a textbook.
References (in ../../references/):
decomposition-rules.md— hierarchy, sizing (1 Story = 1 sprint = 1 PR), story-point heuristic, DoR.ticket-structure.md— draft format + draft file constraints (frontmatter, filename regex).azure-mechanics.md— create/link MCP calls + the two linking gotchas + rendering rules.audit-checklist.md— fidelity / coverage / DoR postflight.
References (skill-specific, in ./references/):
../../common/templates/canonical-user-story.md— read-only shape contract for enriched Story drafts (seven emoji sections perticket-structure.md, pt-BR labels). Do not edit; validate every draft against this template whenlanguageispt-BR.../../common/contracts/decompose-backlog/canonical-user-story.en.md— same shape contract with English section labels. Use whenlanguageisen.../generate-plain-language-documentation/references/integration-notes.md— prose polish sub-pass in DRAFT and ENRICH phases.
Input
A parent work item id (Epic or Feature). Optional: target iteration, story-point ceiling override,
language (en | pt-BR, default pt-BR) — selects which canonical template and section-label
set (ticket-structure.md → Body sections) every drafted Story uses; set it once for the whole run
and record it as language: in each draft's frontmatter so validate-artifact checks against the
matching labels.
Phases
1. INGEST
Read the parent via wit_get_work_item (expand relations). Capture the original text VERBATIM, its
acceptance criteria, and the parent chain. Read any linked spike/wiki. Determine parent type.
Branch: if the parent is an Epic, STOP and ask whether to create intervening Feature(s) first
or target an existing child Feature (see decomposition-rules.md → Parent-type branch). Stories never
attach to an Epic.
2. DECOMPOSE
Apply the sizing rule and story-point heuristic from decomposition-rules.md. Produce an ordered list
of Story stubs (title + one-line scope + dependencies), each tracing to a verbatim slice of the parent.
── GATE 1 — present the split and WAIT for explicit approval before drafting anything.
3. DRAFT
Per approved stub, write a local draft per ticket-structure.md and the canonical template matching
the run's language (../../common/templates/canonical-user-story.md for pt-BR,
canonical-user-story.en.md for en) — hook-valid frontmatter (type, no status, language: set
to the chosen value), filename regex with the Feature-id prefix, the 7 body sections in canonical
order using that language's labels. Content hygiene applies.
Plain-language sub-pass: Read
../generate-plain-language-documentation/references/integration-notes.md § decompose-backlog; run a
generate-plain-language-documentation pass on scope lines and section prose (glossary-verify via
../generate-plain-language-documentation/references/assets/tech-glossary-en-pt-br.json when locale
is pt-BR).
4. ENRICH
Tighten each draft to the team format: WHAT not HOW, ASCII diagrams, de-dup (each fact once), story-point justification with the per-driver MAX. The enriched body IS the exact Azure description.
Plain-language sub-pass: Read
../generate-plain-language-documentation/references/integration-notes.md § decompose-backlog; polish
narrative paragraphs inside sections (not emoji headings; not story-point driver tables).
Glossary-verify via ../generate-plain-language-documentation/references/assets/tech-glossary-en-pt-br.json
when locale is pt-BR.
── GATE 2 — show the final body and WAIT for thumbs-up before any Azure write.
5. CREATE
Per ../../common/providers.md, create each item with Markdown and its Feature as the immediate
parent. Azure DevOps uses native work-item types; Linear uses the agile:user-story label.
6. VERIFY (structural)
Read each created item back; assert the immediate parent and the Epic→Feature→Story chain.
Reconcile frontmatter with the assigned id (rename file, set provider and provider_id). A failed
assertion STOPS the run.
7. AUDIT (content + coverage)
Run audit-checklist.md: retrieve each item FRESH from Azure; check fidelity, build the
parent-requirement → Story coverage map (flag orphans and scope creep), confirm DoR. Emit the coverage
report. Any gap STOPS and reports — no silent patching.
Operating rules
- Two hard gates (after DECOMPOSE, after ENRICH). Never write to the artifacts path or Azure without the matching approval.
- Every Azure-mutating step is followed by a read-back assertion.
- If the host repo keeps a session artifacts path, write a checkpoint after CREATE and after AUDIT.