Levelup Skill

Summarise what was found:

Artefact chain read for [story-slug]:

• Discovery: [one sentence problem summary]
• Decisions logged: [n]
• ACs shipped: [n/n]
• NFRs verified: [list]
• Standards injected at DoR: [list or "none"]

Ready to scan for learnings. Reply: go

Step 2 — Identify patterns and decisions

Scan across the artefact chain and identify items in these four categories:

Category A — Technical patterns worth standardising

Things that worked well and should be repeated in future stories:

• Naming conventions that emerged
• Architectural patterns adopted
• Test strategies that gave good coverage cheaply
• Error handling approaches proven effective

Category B — Standards gaps exposed

Moments where a missing standard caused confusion, a PR comment, or rework:

• "We had to discuss whether to..."
• "The reviewer asked us to change..."
• "We weren't sure if we should..."

Category C — Reusable decisions

Decisions that are likely to recur and shouldn't be re-litigated each time:

• Technology choices with explicit reasoning
• Scope boundary calls that apply to similar future work
• Risk acceptances that define the team's operating posture

Category D — Anti-patterns to avoid

Methods tried and found to not work:

• Approaches abandoned mid-implementation
• Test strategies that gave false confidence
• Patterns that caused rework

Present findings for confirmation:

Patterns identified:

[Category A — n items]

• [summarise each]

[Category B — n items]

• [summarise each]

[Category C — n items]

• [summarise each]

[Category D — n items]

• [summarise each]

Do you want to capture all of these, or select specific ones? Reply: all — or list which to keep

Step 3 — Classify and route

For each accepted item, determine where it should be written:

Confirm routing before writing:

Proposed writes:

1. [item]: → .github/standards/api/api-design.md (new rule under "Naming")
2. [item]: → .github/architecture-guardrails.md (new ADR: "Adopt [pattern]")
3. [item]: → artefacts/[feature]/decisions.md (recurring scope call)

Confirm these destinations? Reply: yes — or adjust any routing

Step 4 — Write the updates

For each confirmed item, produce the exact text to add and the target location.

Standards files: append under the relevant section heading. Create the file and section if they don't exist — don't leave orphaned rules.

Architecture guardrails: append a new ADR block:

### ADR-[n]: [Decision title]
**Date:** [YYYY-MM-DD]
**Status:** Active
**Context:** [one sentence — what situation prompted this]
**Decision:** [what was decided]
**Rationale:** [why]
**Consequences:** [what this rules in or rules out going forward]

copilot-instructions.md: if a coding standard is universal (not domain-specific), add it to the coding standards section.

Confirm each write before creating it:

Writing to .github/standards/api/api-design.md: [preview of exact text]

Confirm? Reply: yes — or adjust the wording

Step 5 — Update standards index

After writing to any standards file:

1. Open .github/standards/index.yml
2. Verify the file is listed under the correct domain
3. If not listed — add it

Standards index updated: [file] added to [domain] section.

Completion output

✅ Levelup complete — [story-slug]

Patterns extracted: [n] Standards updated: [list of files] ADRs added: [n] Decisions logged: [n]

These learnings are now active — /definition-of-ready will inject them into future stories with matching domain tags.

Recommended: commit .github/standards/ and .github/architecture-guardrails.md with message: chore: extract learnings from [story-slug]

What this skill does NOT do

• Does not reopen the story or test plan — artefacts are read-only post-merge
• Does not create new stories from the learnings — that is a future /discovery or /definition run
• Does not update the story or DoD retroactively
• Does not assign action items to people — it writes to files only

State update — mandatory final step

Mandatory. Do not close this skill without completing this write.

Update .github/pipeline-state.json:

• Under the relevant feature: set lastLevelup: "[YYYY-MM-DD]"
• Under standards: add each updated file name to a recentUpdates list

Confirm the write in your closing message: "Pipeline state updated ✅."