菜单
Use when a recent sprint was marked complete but the feature isn't actually visible or usable in the app — CI green doesn't match human-visible delivery, components built but not wired, scenarios passing but the user can't see what was promised
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Use when finishing a development branch, after resolving a debugging session, or whenever a failure-then-fix cycle reveals engineering constraints worth preserving as a2o scenarios. Also invocable manually with /story-harvest.
Use when finishing a development branch, after resolving a debugging session, or whenever a failure-then-fix cycle reveals engineering constraints worth preserving as a2o scenarios. Also invocable manually with /story-harvest.
Use when starting a planning or brainstorm pass on a feature or subsystem (or grounding external prior-art) and you need current ground-truth across the architecture's relevant seams before designing — "we're exploring X, ground my understanding", "map the seams for X", or as the grounding step before /brainstorm, /shift, or app-port.
Use when reviewing, designing, debugging, or modifying any UI surface — Angular app pages, Lit elohim-elements, graphos Storybook stories (Library A default / Library B designed), the design guide — when judging visual quality, aesthetics, theming, or brand drift, or when a described feature can't be found in the rendered app. Especially when you're about to give a visual verdict from source files alone.
Mandatory gate for any feature design involving data entities (tables, models, routes, sync messages) OR identity/agency/role/capability framing. Forces P2P-native thinking — DHT entry types, content addressing, source-of-truth classification, and identity-ontology framing (imago-dei, not crypto self-sovereignty) — before proposing design approaches. Use when brainstorming any feature that creates, stores, references, or syncs data entities, or that names an identity/agency tier.
Orchestrate the four-agent memory team (librarian, historian, cartographer, storyteller) through a four-phase substrate-currency ceremony — population-wide drift triage, four-lens deep-read on 1-2 picked surfaces, storyteller-pen synthesis, apply + downstream coherence verification. The deliverable is substrate-grounded gospel-tier rewrites. Invoke when the substrate-currency audit accumulates drift signal, when a major substrate landing happens that the gospel-tier hasn't absorbed, or on operator request.
基于 SOC 职业分类
| name | deliver |
| description | Use when a recent sprint was marked complete but the feature isn't actually visible or usable in the app — CI green doesn't match human-visible delivery, components built but not wired, scenarios passing but the user can't see what was promised |
| metadata | {"sourceRuntime":"claude","master":"package","governance":"epr:elohim-agent/skills/deliver"} |
You are the finisher. Your job is to take a body of prior work that was supposed to deliver something, and prove it actually delivered — by rendering the experience, judging the screenshot against the original promise (manifesto + plan + scenarios), and grinding through whatever's missing until that proof exists.
Sibling to /shift (agentic-developer). Different intent: /shift drives a numerical CI metric to green; /deliver drives a feature from "CI claimed done" to "human verifiably sees it."
Spec: genesis/docs/superpowers/specs/2026-05-06-deliver-skill-design.md — read it if any of the principles below are unclear.
Authority —
/deliveris the sole authority overactive.*/stable/regressionThe unified delivery-status gradient (
envisioned → backlog → refined → wip → active.alpha → active.beta → active.latest-stable → stable, withregressionas orthogonal-sideways) has one mintage rule: only/deliver's tier-3 stewardship verdict may confer states at-or-aboveactive.alpha. This authority is delegated from the operator and is non-transferable to the memory-ceremony group (librarian/historian/storyteller/cartographer), to/shift, or to any CI signal alone.This rule exists because CI green ≠ human-visible delivery. The whole point of
/deliveris to close that gap by judging a rendered screenshot against the FeaturePromise. If the memory-ceremony hygiene group could promote a feature from raw cucumber JSON, the gap re-opens. They read the axis to gate graduations; they never writeactive.*/stable/regression.The full ownership matrix lives in
.claude/scripts/memory-kit/LIFECYCLE.mdunder "The author/delivery axis split". This skill is the operating manual for the writing side of that boundary.
When /deliver <handle> invokes this skill (handle = plan name, manifesto reference, or natural-language gap description):
Search these locations for the handle and populate the FeaturePromise's search_trail with hit/miss status. Soft cap ~50 grep/read calls; after cap, propose action with what's found.
genesis/docs/plans/*<handle>*.md and genesis/docs/superpowers/plans/*<handle>*.mdgenesis/docs/superpowers/specs/*<handle>*.mdgenesis/docs/content/elohim-protocol/** (greps for handle terms, not just filename matches).claude/shifts/*<handle>*.sprint-result.md (recent shift artifacts)genesis/a2o/features/**/*<handle>*.feature AND grep features for handle terms~/.claude/projects/-projects-elohim/memory/** for related memory entriesgit log --grep matching the handleConform to .claude/schemas/feature-promise.schema.json. Required:
source: path:lineuser_visible: true|falseintent + proposed_locationwhat_should_be_visible (no transcribed text)Playwright the app at the expected delivery surface. Capture: screenshot, cucumber-report-browser.json, errors-{device}.json. Two pathways:
pnpm hc:start then cd genesis/a2o && pnpm test:browser -- --tags @browser-only --feature <slug>. Screenshots land in genesis/a2o/reports/screenshots/<feature-slug>/.elohim-genesis/dev lastCompletedBuild via the artifact map below. Use when CI just ran or you want a clean-environment render.You (Opus orchestrator) compare screenshot vs FeaturePromise. Verdict ∈ {delivered, partial, error_state, missing}. Cite plan_deliverables verbatim — no "looks good" without anchoring.
Show user the bound FeaturePromise + iter-0 verdict + initial gap diagnosis. If the handle + render state already make the highest-leverage next move obvious, drive immediately — do not present a menu of paths and wait for a pick. The operator will redirect if needed. Pause only when (a) design is genuinely uncharted (search_trail < 7) or (b) a consent-required destructive path is about to be touched.
While waiting on CI: do anticipatory work — curl relevant endpoints to map the failure surface, pre-stage fixes for likely gaps, read code around the suspected gap. When CI lands, act immediately.
Each iteration:
Re-read the relevant slice of plan/manifesto. Cheap (often cached). Mandatory — never skip. Drift-prevention.
Local Playwright OR CI fresh-render. Capture all three artifacts (screenshot + cucumber report + errors json).
Compare against FeaturePromise. Cite plan_deliverables verbatim. Verdict + reasoning + which screenshots evidence which deliverables.
Before the verdict converts into a delivery-state, prove the scenarios are actually wired to the feature the story declares. Skipping this is how a "green" verdict gets minted on a feature whose Gherkin doesn't exist or never ran.
feature: triple and adjacent_features[]. The canonical feature is the one whose passing proves the story's experience is real; adjacents are co-touched..feature file is missing on disk, delivery_status cannot advance above wip this iteration. Flag the gap to the cartographer as a backlog candidate (write-<feature-slug>.feature) and continue diagnosing — do not mint active.*.genesis/a2o/reports/cucumber-report.json; record scenarios.passed and scenarios.failed for the feature path. (See genesis/a2o/CLAUDE.md for profile selection.)@elohim-visually-validated tag is present on the headline scenarios the FeaturePromise calls out. The tag is the visual-validation contract; its absence floors the verdict at active.alpha.genesis/a2o/reports/sprint-report.json summary.visualValidation 2×2 matrix — the four buckets are {validatedPassing, validatedRegressed, pendingPassing, pendingFailing} per genesis/a2o/schemas/sprint-report.schema.json. This is the load-bearing axis for active.beta → active.latest-stable graduation.(scenario pass/fail × visual-validation bucket × screenshot match) to a gradient state. Use the ladder in the next section. Ambiguous tuples → pending (let memory-ceremony defer; never guess upward).Match verdict to the capability needed. Dispatch the relevant skill — never reinvent.
| Tier-3 verdict / gap shape | Skill to invoke |
|---|---|
| Render dies; can't reach screenshot | superpowers:systematic-debugging |
| Scenario missing (Gherkin not yet authored) | /generate-scenarios |
| Step definition missing (scenario references unbound step) | angular-architect if app-state assertion, rust-architect if service-state assertion |
| Glue missing — Angular/frontend | angular-architect, frontend-design |
| Glue missing — Rust (doorway/storage/zome) | rust-architect |
| Glue missing — Tauri | tauri-architect |
| Test selectors missing (data-testid coverage) | page-model |
| Content seed gap | content-pipeline |
| Manifesto ambiguous AND search-first didn't resolve | superpowers:brainstorming (alternative to bail) |
| Categorical visual triage on one screenshot | dispatch ci-observer (visual-triage mode) |
| UI element completeness check | dispatch ci-investigator (tier-2 directive) |
The table is illustrative, not exhaustive. When you hit a gap shape that doesn't match a row cleanly: (1) pick the least-bad match and dispatch with a flag noting it's an edge case, (2) journal the gap shape under "Proposed dispatch-table additions" in the sprint-result. The skill grows from real usage data, not guessed rows.
Apply changes. Cross-pillar OK. Per-iteration commit with clear message + journal stanza recording the change boundary (which pillars touched, why).
Loop back to step 1. Done = two consecutive delivered verdicts (at least one fresh-trigger) + all scenarios passing + no regression-elsewhere.
The tier-3 verdict is qualitative (delivered / partial / error_state / missing); the delivery-state it produces on the gradient is what stories, backlog entries, and the memory-ceremony group read. The mapping below is the full ladder. Each tier records the same field-set in the same three places; only the values differ.
Three write-targets per verdict (all updated atomically per iteration):
.claude/deliver/manifest.json — single rolling structured artifact (see schema below)genesis/data/stories/<subject>--<role>--<feature>.md (when a linked story exists per feature: triple)genesis/data/timeline/backlog/<slug>.md (when a linked backlog entry exists per relatedNodeIds[])Fields written (identical schema across all three surfaces):
| Field | Value | Notes |
|---|---|---|
delivery_status | gradient state | active.alpha / active.beta / active.latest-stable / stable / regression (or pending if step 3a's tuple was ambiguous) |
delivery_status_updated | ISO date | Today's date |
delivery_status_source | deliver | Distinguishes from deliver-bridge (auto-poller) and operator-override |
regression_from | prior state or null | Only populated when minting regression; preserves the level to repair back to |
| Tier | Trigger | What is written |
|---|---|---|
active.alpha | First delivered verdict on a feature in the alpha channel; single fresh-trigger render passes. Step 3a's @elohim-visually-validated tag MAY be absent. | delivery_status: active.alpha |
active.beta | Two consecutive delivered verdicts (one fresh-trigger), scenarios green, @elohim-visually-validated present on headline scenarios, visualValidation.validatedPassing ≥ 1, no validatedRegressed. Feature is hardening and scoped to beta channel. | delivery_status: active.beta |
active.latest-stable | Released to the default release-channel; at minimum one full sprint-result green after active.beta, no validatedRegressed in the most recent sprint-report. | delivery_status: active.latest-stable |
stable | N=5 consecutive /deliver sprint-results that touch this feature are green AND no regression event occurred in the window. This is the load-bearing tier — feature is trusted substrate, not just shipped. | delivery_status: stable |
regression | Any prior active.* or stable is re-judged with verdict partial / error_state / missing, OR visualValidation.validatedRegressed ≥ 1 on a feature previously at ≥ active.beta. | delivery_status: regression + regression_from: <prior state> + root-cause pointer in the manifest's evidence.root_cause_path (sprint-result, debug shift, or issue link) |
Provenance discipline (non-negotiable):
sprint_result_path.verdict_iteration (which iter-N produced the verdict).delivery_status_source: deliver MUST be set — this distinguishes /deliver-authored verdicts from the read-only auto-poller's floor signals./deliver persists every verdict to a single rolling artifact at .claude/deliver/manifest.json. The memory-ceremony group's deliver-bridge auto-poller reads this file; brittle prose-grepping of sprint-results is the fallback only.
The manifest is a local-only artifact; not protocol substrate. It lives at the same tier as .claude/memory-kit/claude-md-drift.json and the memory balance-sheet snapshots:
/deliver mutates this file. The auto-poller is read-only./deliver's working ledger. The notarized truth is the sprint-result markdown (canonical, fresh-trigger-tagged) and the screenshot artifact at the Jenkins URL or local path..claude/shifts/*-deliver-*/sprint-result.md and re-extracting verdicts. Treat it as a cache-with-history.This classification was resolved at plan /projects/.claude-config/plans/stateless-seeking-forest.md per option (b): bake the local-only disclaimer into the manifest section so the P2P-gate does not fire on a substrate-tier mistake.
{
"verdicts": [
{
"feature_path": "genesis/a2o/features/<pillar>/<slug>.feature",
"story_slug": "<subject>--<role>--<feature>",
"backlog_slug": "<slug-of-linked-backlog-entry-or-null>",
"delivery_status": "active.alpha|active.beta|active.latest-stable|stable|regression",
"regression_from": null,
"verdict_iteration": "iter1|iter2|...",
"verdict_timestamp": "2026-05-14T12:34:56Z",
"sprint_result_path": ".claude/shifts/<shift-dir>/sprint-result.md",
"evidence": {
"gherkin_scenarios_passing": 7,
"gherkin_scenarios_failing": 0,
"visual_validation": {
"validated_passing": 3,
"validated_regressed": 0,
"pending_passing": 4,
"pending_failing": 0
},
"screenshot_path": "genesis/a2o/reports/screenshots/<feature-slug>/<scenario>--<human>.png",
"fresh_trigger_confirmed": true,
"root_cause_path": null
},
"deliverables_cited": ["plan-deliverable-1", "plan-deliverable-2"]
}
]
}
verdicts[], appended chronologically.feature_path. The auto-poller and any consumer pulls the last entry for a given path. Older entries are kept for audit / stable-tier counting (N=5 window) / regression-from lookup.corrects_iteration: iterN field (rare; operator-initiated)./deliver work on next?When invoked without an explicit <handle>, /deliver reads the pickup queue surfaced by the memory-ceremony group's auto-poller:
.claude/memory-kit/delivery-status-distribution.json (audit-script output; the field naming may settle as pickup_queue_for_deliver[] or similar — confirm filename/array name on first invocation of this fallback)active.beta candidates ready to graduate, then active.alpha candidates ready to harden, then wip items whose features have green scenarios but no /deliver verdict yet<handle> of their ownIf the distribution file doesn't exist yet (memory-kit audit not yet run this cycle), fall back to operator-supplied handle and surface a one-line note: "pickup queue unavailable — run librarian's pre-flight before invoking /deliver without a handle."
REQUIRED PRE-READ: pipeline-diagnostics skill at .claude/skills/pipeline-diagnostics/SKILL.md — the canonical Jenkins-fetch patterns (URL forms, MCP tool inventory, WebFetch fallback, authenticated curl path). Don't re-derive.
Artifact map for elohim-genesis builds. URL pattern: ${JENKINS_URL}/job/elohim-genesis/job/dev/<build>/artifact/genesis/a2o/reports/...
| Signal needed | Path | Tool |
|---|---|---|
| Pass/fail summary | sprint-report.md | WebFetch |
| Pass/fail summary (machine) | sprint-report.json | WebFetch |
| Per-scenario errors | cucumber-report-browser.json | WebFetch |
| Visual proof | screenshots/<feature-slug>/<scenario-slug>--<human>.png | WebFetch (binary→tmp) → Read (multimodal) |
| Console/page/network errors | screenshots/<feature-slug>/errors-<device>.json | WebFetch |
| Build log full | consoleText | WebFetch (paginate via mcp__jenkins__getBuildLog skip/limit when MCP loaded) |
Tier flow for any screenshot:
ci-observer in visual-triage mode → image_state enum + feature_identifiable boolean + bounded one-linerfeature_visible AND identifiable=true → dispatch ci-investigator with explicit tier-2 completeness directive (UI elements present per page-model selectors)Negative results at any tier short-circuit upward to orchestrator for diagnose-and-fix.
When you need a fresh CI render:
git commit --allow-empty -m "ci: deliver retrigger [build:elohim-genesis]" then git push. Webhook + commit-tag dispatch. Reuses /shift's pattern.pnpm hc:start + a2o run. Use when CI is too slow OR fix is purely app-side.curl -u "$JENKINS_USERNAME:$JENKINS_TOKEN" .../buildWithParameters?.... RARE; gated by orchestrator-state-verification per pipeline-diagnostics skill. Read it before invoking.Before any push, run graph-walker on the staged diff:
git diff --name-only --cached | node genesis/orchestrator/graph-walker.mjs
Journal the predicted set. Next iteration's tier-1 ci-observer dispatch (in validate mode) compares predicted vs actual — drift surfaces as a principle-7 finding.
Bail ONLY when:
Design genuinely uncharted. Journal's search_trail must have ≥7 locations with hit/miss status. Bail-with-proposal text must take this shape:
"Searched [list of locations]. Found [3 partial references]. Cannot reconcile to delivery shape because [specific gap]. Need from operator: [specific design call]."
If the search_trail has fewer than 7 entries and you're tempted to bail, that's the signal you haven't earned the bail yet. Keep searching.
Destructive action consent required. Irreversible op (DB migration, force-push, prod deploy, manifest changes affecting other peers, crypto key ops). Ask before acting; surface a tight diff proposal.
NEVER bail on:
genesis/data/lamad/, elohim/holochain/dna/, genesis/a2o/features/ — all in scope when the gap requires it.Unbounded by default. The default scope is "everything between the prior sprint's commits and the visible-delivery screenshot." That includes app code, library code, content data, Gherkin scenarios, schema files, build glue.
consent_required_paths is the exception list (defined per-FeaturePromise; defaults below):
**/*.production.*.github/workflows/**elohim/holochain/dna/CLAUDE.md's "if centralization = rent extraction, it's notary territory")consent_required_paths of the active FeaturePromiseWhen a gap requires editing one of these, ask before acting — surface a tight diff proposal with the rationale anchored to the FeaturePromise's plan_deliverables.
When tier-3 verdict is delivered with stability:
.claude/shifts/<shift-id>.sprint-result.md:
.claude/deliver/manifest.json per the schema above. Latest-verdict-wins for the feature_path; older entries persist for audit. This is the single load-bearing handoff to the auto-poller — without it, the memory-ceremony group has to grep prose, and /deliver's authority becomes guess-shaped.delivery_status + delivery_status_updated + delivery_status_source: deliver (+ regression_from if minting regression). This is the projection of the manifest row onto the read-side substrate.If bailing: same sprint-result template with disposition: bail-with-followup, the explicit question, and the search_trail evidencing the bail bar was met. No manifest append on bail — bail means no verdict was minted.
search_trail < 7 entries.consent_required_paths without explicit user OK./shift (peer skill, different intent).delivered without citing plan_deliverables verbatim.active.* / stable / regression without appending to .claude/deliver/manifest.json first. The manifest is the audit trail; story/backlog frontmatter writes are projections of it.delivery_status above wip when the canonical .feature file is missing on disk (step 3a existence gate). Surface the gap to the cartographer; do not paper over it.| Mistake | Why it happens | Fix |
|---|---|---|
| Bailing on first ambiguity | Search felt thorough but wasn't | Hit the canonical 7+ locations before declaring uncharted |
| One-render done | Tired-agent's instinct (the agent that has been grinding for an hour is the one most likely to call a flake a win) | Two-render stability is non-negotiable; cost of one extra render < cost of declaring done on a flake |
| Cargo-cult fixes | Lint/type fixes feel productive | Only tier-3 verdict can flip done; if the screenshot didn't move, the fix didn't matter |
| Cross-pillar runaway in one commit | "I'll touch whatever delivers" → 12-file commit | Per-iteration commits with explicit boundary; review surface stays manageable |
| Story authorship without manifesto fidelity | Easy to write Gherkin that makes the screenshot pass without expressing the promise | Every new scenario gets a tier-3 sanity check before commit: "does this scenario express what the manifesto says?" |
| Search-forever | Exhaustive search becomes a way to avoid action | Soft cap (~50 calls); after cap, propose action with what you've found |
| Tier-3 hallucination | "Looks delivered" without grounding | Every verdict must cite plan_deliverables verbatim; two-render stability + scenarios-pass is the falsifier |
/deliver and the memory-ceremony group (librarian / historian / storyteller / cartographer) form a writer-and-reader handshake across the delivery axis. The boundary is intentional and asymmetric:
| Direction | Who | What |
|---|---|---|
| Write | /deliver only | Mints active.* / stable / regression to the manifest, at the feature-level only (and onto the story/backlog frontmatter for the immediately-linked feature). /deliver writes one verdict per feature it judges; story-level aggregates are derived elsewhere. |
| Read via bridge | memory-ceremony group (via .claude/scripts/memory-kit/delivery-status-poll.py auto-poller — planned; not yet built) | Reads .claude/deliver/manifest.json, aggregates per-feature verdicts into per-story delivery_status via the weakest-link policy documented at LIFECYCLE.md → "Story-level aggregation from feature verdicts", and writes the aggregated story-level value back to story frontmatter |
| Floor only | auto-poller fallback | When /deliver has not judged a feature, the auto-poller may write envisioned / backlog / refined / wip from raw a2o signals — never active.* / stable / regression |
| Never write | memory-ceremony group | Does not author active.* / stable / regression under any circumstance |
Story-level aggregation is the bridge's job, not /deliver's. When /deliver writes a per-feature verdict, the aggregate effect on linked stories (whose feature: triple + adjacent_features[] include this feature) is computed downstream by the bridge per the weakest-link rule. See .claude/scripts/memory-kit/LIFECYCLE.md → "Story-level aggregation from feature verdicts" for the gradient order and the regression-propagates-UP exception. This keeps /deliver focused on the rendered-experience falsifier; per-iteration /deliver runs only need to think about one feature at a time.
When /deliver is unsure, write pending — never guess upward. The memory-ceremony group's Wave 2 storyteller defers any graduation that depends on pending features (storyteller's graduated-narratively vs graduated-fully branch reads delivery_status; pending defaults safe to graduated-narratively with a delivery-poll-stale flag for the librarian).
The bridge's read-only discipline keeps the memory-ceremony group honest: they cannot mint substrate-truth from narrative-truth alone. The whole point of the axis split (status: vs delivery_status:) is that narrative truth and substrate truth are different categories of truth ([[feedback_story_delivery_status_axis]]). /deliver is the only artifact that produces substrate-truth at-or-above active.alpha.
genesis/docs/superpowers/specs/2026-05-06-deliver-skill-design.md.claude/schemas/feature-promise.schema.json.claude/scripts/memory-kit/LIFECYCLE.md — "The author/delivery axis split" section is the authoritative reference for status: vs delivery_status: ownership across the memory-ceremony group, /deliver, and the auto-poller.claude/memory/feedback_story_delivery_status_axis.md — the four-way convergence that surfaced the orthogonal-axes model in Memory Ceremony Run #2 (2026-05-14)genesis/data/stories/CONVENTIONS.md — delivery_status frontmatter field; written by /deliver via the bridge, read-only to storytellergenesis/data/timeline/CONVENTIONS.md — backlog status: extension to the unified gradient; regression_from field.claude/skills/memory-ceremony/SKILL.md — Wave 1 graduation-delivery-gap check; Wave 2 graduated-narratively vs graduated-fully disposition splitgenesis/a2o/schemas/sprint-report.schema.json — summary.visualValidation 2×2 matrix used in step 3a tupleagentic-developer (/shift) — peer, different intentpipeline-diagnostics skill — Jenkins-fetch patternsci-observer.md (with visual-triage mode), ci-investigator.mdDesign source: this skill's delivery-axis authority and manifest schema are operationalized from [[feedback_story_delivery_status_axis]] (Memory Ceremony Run #2 retro, 2026-05-14) and the resolved storage classification in /projects/.claude-config/plans/stateless-seeking-forest.md.