Stats
Actions
Tags
From FOUNDRY — Value-Flow Production System
FOUNDRY — Technical Tooling for Software Evolution. The production orchestrator for this production system. Ingests a complete ROADMAP.md, tiers all items by PRIORITY_STATUS against a token budget, decomposes the full backlog into parallelised expert subagents (PHASE_POOL + VALUE_HANDLER_POOL), drives every item through the full DEV_SYSTEM pipeline (EARS → FEATURE → TEST → IMPLEMENT → STORY), and records every completion in IDEA_COST.jsonl. FOUNDRY is the engine of the idea-to-production conveyor. Trigger when the user says: "run the foundry", "process the roadmap", "orchestrate the roadmap", "ship the backlog", "build everything", "start FOUNDRY", "run a foundry cycle", "what would this cost?", "estimate the backlog", or "inspect FOUNDRY".
How this skill is triggered — by the user, by Claude, or both
Slash command
/foundry:builderThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
> *Technical Tooling for Software Evolution*
Technical Tooling for Software Evolution
FOUNDRY is the production orchestrator of the idea-to-production conveyor. It takes a fully
populated ROADMAP.md, tiers all pending items by token budget and priority,
decomposes the work into parallelised specialist subagents, drives every item
through the complete DEV_SYSTEM pipeline, and records the cost of every
completion in IDEA_COST.jsonl so the system gets smarter with every cycle.
FOUNDRY orchestrates. It does not implement.
FOUNDRY spawns agents, sequences phases, invokes reviewers, and records costs. It does not write EARS statements, Gherkin, test code, or production code directly. All artefact production is delegated to PHASE_POOL and VALUE_HANDLER_POOL agents. FOUNDRY's only direct writes are to
doc/FOUNDRY_PLAN.mdandIDEA_COST.jsonl.CODE_QUALITY is the knowledge-home for all code design and review decisions. FOUNDRY defers to it at every implementation and review stage.
Token scarcity is a first-class constraint. Every orchestration decision must account for token budget. Tiering, parallelisation, and batching are all in service of maximising FEATURE value per token spent.
KAIZEN applies to all documents, agents, and artefacts produced by FOUNDRY. The covenant in
${CLAUDE_PLUGIN_ROOT}/knowledge/architecture/kaizen-covenant.mdtravels with everything.
| User says / does | Entry point |
|---|---|
| "run the foundry" / "process the roadmap" / "orchestrate" | → §3 ROADMAP INGESTION |
| "ship the backlog" / "build everything" / "start FOUNDRY" | → §3 ROADMAP INGESTION |
| "run a foundry cycle" / "full send" | → §3 ROADMAP INGESTION |
| "what would this cost?" / "estimate the backlog" | → §4.3 TOKEN ESTIMATION only |
| "inspect FOUNDRY" / "run the inspector" | → §13 INSPECTION PROTOCOL |
| "add to FOUNDRY" / "update FOUNDRY knowledge" | → §14 SELF-IMPROVEMENT |
| "what tier is [item]?" / "reprioritise" | → §4 TIER ASSIGNMENT only |
Before any processing, silently check:
ROADMAP.md or doc/ROADMAP.md. If absent, stop and tell the user:
"No ROADMAP.md found. FOUNDRY requires a populated roadmap. Run IDEATOR then ROADMAPPER to create one, then return here."
IDEA_COST.jsonl (project root or doc/) if it exists — extract
average token cost per item type for budget estimation.package.json, pyproject.toml,
Cargo.toml, go.mod, dominant file extensions. Note languages and
frameworks — this informs VALUE_HANDLER_POOL selection.doc/SUBJECT_MATTER_UNDERSTANDING.md. If absent, note it — LEAD
ENGINEER will create it during §5.doc/SPECIFICATION.ears.md and features/ directory.
Note what already exists — completed items should not be reprocessed.Record findings. Narrate only if something requires user attention (e.g., no roadmap, or a conflict between existing artefacts and roadmap status).
Read the full roadmap. For each entry extract:
ID, TITLE, STATUS, PRIORITY_STATUS, BRIEF, SLUG
Filter: Only process items with STATUS = PENDING or BACKLOG.
Skip: COMPLETE, DEFERRED, IN PROGRESS (unless user explicitly says to resume).
Present the filtered list to the user:
FOUNDRY — Roadmap Ingestion
────────────────────────────
Found N items eligible for processing:
#1 [TITLE] PRIORITY: HIGH Est. tokens: ~N
#3 [TITLE] PRIORITY: MEDIUM Est. tokens: ~N
#7 [TITLE] PRIORITY: LOW Est. tokens: ~N
...
Proposed tier structure: [see §4]
Proceed with full cycle, or adjust tiers first?
Wait for user confirmation or adjustments before proceeding to §5.
See ${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/tier-assignment.md for the full assignment matrix.
| PRIORITY_STATUS | Default tier | Override condition |
|---|---|---|
CRITICAL | PRIMARY | — |
HIGH | PRIMARY | Unless token budget exhausted |
MEDIUM | SECONDARY | May promote to PRIMARY if it unblocks HIGH items |
LOW | TERTIARY | — |
BACKLOG | TERTIARY+ | Processed only when all higher tiers complete |
Items with shared infrastructure dependencies are promoted to the earliest tier that needs them — a MEDIUM item that provides a data layer for three HIGH items moves to PRIMARY.
Each tier has a token budget cap (default: 40% of total estimated session budget for PRIMARY, 35% for SECONDARY, remainder for TERTIARY+). The LEAD ENGINEER may adjust these caps during §5 based on item complexity analysis.
Use IDEA_COST.jsonl history if available. Estimation heuristics when no
history exists:
| Item complexity | Estimated tokens |
|---|---|
| Trivial (1 EARS statement, 1 scenario) | 8k–15k |
| Small (2–4 EARS, 3–6 scenarios) | 15k–35k |
| Medium (5–8 EARS, 6–12 scenarios) | 35k–80k |
| Large (9+ EARS, 12+ scenarios) | 80k–200k |
Record estimates. Compare against actuals in IDEA_COST.jsonl after completion to refine future estimates.
Spawn builder-lead agent with:
LEAD ENGINEER is responsible for producing doc/FOUNDRY_PLAN.md containing:
# FOUNDRY Plan — [project] — [date]
## Stack Manifest
[Languages, frameworks, test runners, BDD tools, E2E tools]
## Subject Matter Understanding
[Domain context carried by all agents — see ${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/subject-matter-understanding.md]
## Shared Infrastructure Map
| Component | Used by items | Build in item | Notes |
|---|---|---|---|
| [e.g. User model] | #3, #5, #7 | #3 | Items #5, #7 depend on #3 |
## Work Decomposition
### Item #N — [TITLE]
- Tier: PRIMARY / SECONDARY / TERTIARY
- Token budget estimate: ~Nk
- Tasks: [ordered list of sub-tasks]
- Dependencies: [item IDs this depends on]
- Parallel-safe with: [item IDs that can run concurrently]
## Parallel Grouping
[Which items run concurrently within each tier]
## Build Order
[Dependency graph across all items]
## VALUE_HANDLER_POOL Required
[List of language/stack agents to spawn: PYTHON-AGENT, JS-AGENT, etc.]
LEAD ENGINEER also creates doc/SUBJECT_MATTER_UNDERSTANDING.md if absent.
See ${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/subject-matter-understanding.md for the template.
Heavy items — delegate the breakdown. When a roadmap item is too large to decompose into atomic vertical slices by hand, the LEAD ENGINEER may spawn the
handler-roadmap-decompositionvalue-handler to produce the INVEST-sized, dependency-ordered, phase-mapped job list that fills the Work Decomposition and Build Order sections above. That handler plans only — it returns the job breakdown; the LEAD ENGINEER remains the orchestrator that sequences and dispatches.
Process tiers in sequence. Within each tier, process items concurrently where the FOUNDRY_PLAN marks them as parallel-safe.
For each TIER in [PRIMARY, SECONDARY, TERTIARY, ...]:
1. Display tier start banner:
══════════════════════════════════════
FOUNDRY — Tier: PRIMARY (N items)
Token budget: ~Nk
══════════════════════════════════════
2. For each ITEM in TIER (parallel where parallel-safe):
→ §7 PHASE_POOL PIPELINE for this item
3. On item completion:
→ §12 IDEA_COST RECORDING for this item
4. After all items in tier complete:
- Compare actual vs estimated token costs
- Adjust next tier estimates accordingly
- Display tier summary before proceeding
Between tiers, pause and report:
Tier [NAME] complete.
Items completed: N
Actual tokens: ~Nk (estimated: ~Nk, delta: ±N%)
Proceed to [NEXT TIER]? (N items, est. ~Nk tokens)
Each roadmap item passes through 6 phases in strict sequence. Between every phase, the REVIEWER panel evaluates the output (§9) before the next phase begins.
Preferred: Use the named DS step agents from ${CLAUDE_PLUGIN_ROOT}/agents/ds-step-*.md
rather than spawning general agents with role prompts. Named agents carry explicit
inputs, outputs, handoff schemas, sentinel emissions, and reviewer rules — they
are single-responsibility and cold-start safe.
The full pipeline and per-item orchestration loop is documented in
${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/orchestration-loop.md. The PHASE_POOL maps to the DS steps as follows:
| FOUNDRY Phase | DS Steps | Named Agents | Model |
|---|---|---|---|
| Pre-cycle (plan) | step-0 | ds-step-0-plan | sonnet (default) |
| EARS-AGENT | step-1 | ds-step-1-ears | opus (EARS) |
| FEATURE-AGENT | step-2 | ds-step-2-feature-docs | opus (stories) |
| TEST-AGENT | step-3 + step-4 | ds-step-3-tests, ds-step-4-first-test-run | haiku (test code) |
| IMPLEMENT-AGENT | step-5 + step-6 | ds-step-5-implementation, ds-step-6-green-run | sonnet (default) |
| STORY-AGENT | step-story | ds-step-story-tests | opus (stories) |
| DELIVERY-AGENT | step-7 + step-8 + step-9 | ds-step-7-sync, ds-step-8-commit-message, ds-step-9-commit-push | sonnet (default) |
| REVIEWER | every gate | reviewer, reviewer | opus (review) |
| ARCHITECT | LEAD §5 or IMPLEMENT §4 boundary | handler-architect | opus (review-class judgement) |
Spawn: ds-step-0-plan
Input: Roadmap item brief + SUBJECT_MATTER_UNDERSTANDING + codebase scan
Output: doc/[FEATURE_SLUG]_PLAN.md with checklist and resumption instructions
Context sentinel on completion:
SENTINEL::PLAN_COMPLETE::ROADMAP-{N}::PASS::{plan_path}
→ No reviewer panel required (LEAD ENGINEER reviewed the item in §5)
Spawn: ds-step-1-ears
Input: Plan + SUBJECT_MATTER_UNDERSTANDING + existing EARS spec
Output: EARS statements appended to doc/SPECIFICATION.ears.md
Test policy anchor: Each EARS statement gets a unique ID (EARS-{NNN})
Context sentinel on completion:
SENTINEL::EARS_COMPLETE::ROADMAP-{N}::PASS::{EARS-042,EARS-043,...}
→ REVIEWER panel (EARS-REVIEWER, SMU-REVIEWER) evaluates before Phase 2
Spawn: ds-step-2-feature-docs
Input: EARS statements + sentinel from Phase 1 + SUBJECT_MATTER_UNDERSTANDING
Output: Gherkin scenarios in features/{slug}.feature
Required scenarios per EARS statement:
Each scenario tagged @EARS-{ID}.
Context sentinel on completion:
SENTINEL::FEATURE_COMPLETE::ROADMAP-{N}::PASS::{scenario_count}::{feature_file_path}
→ REVIEWER panel (BDD-REVIEWER, COVERAGE-REVIEWER) evaluates before Phase 3
Spawn: ds-step-3-tests then ds-step-4-first-test-run + appropriate VALUE_HANDLER(s)
Input: EARS statements + feature file + sentinels from Phases 1–2
Output: Failing tests (unit + integration; BDD harness as appropriate) + gap map
Constraint: Tests must be RED at end of this phase. No implementation yet.
Context sentinels on completion:
SENTINEL::TESTS_WRITTEN::ROADMAP-{N}::RED::{test_count}::{ear_ids_covered}
SENTINEL::GAP_MAP_COMPLETE::ROADMAP-{N}::RED::{failing_count}::{ears_ids_exposed}
→ REVIEWER panel (TEST-DESIGN-REVIEWER, COVERAGE-REVIEWER) evaluates before Phase 4
Spawn: builder-lead delegates to VALUE_HANDLER_POOL agents per stack
Input: All previous sentinels + test files + FOUNDRY_PLAN shared-infra map
Primary reference: CODE_QUALITY skill — consulted for all design decisions
Output: Production code making all Phase 3 tests pass
Constraint: Do not modify test files. If a test is wrong, return to Phase 3.
Context sentinel on completion:
SENTINEL::IMPL_COMPLETE::ROADMAP-{N}::GREEN::{files_changed}::{line_coverage_pct}
→ REVIEWER panel (DESIGN-REVIEWER, COVERAGE-REVIEWER) evaluates before Phase 5
Spawn: ds-step-story-tests + interface-appropriate VALUE_HANDLER(s)
Prerequisite: IMPL_COMPLETE sentinel must be present — story tests run only against working code.
Input:
SENTINEL::IMPL_COMPLETE (confirms unit + integration tests green)Medium (case-by-case): browser tests, terminal tests, E2E, or API journey tests — determined by the feature's interface type. Soft rule: more tests means more defects found. Spawn PLAYWRIGHT-AGENT for browser features; use CLI/subprocess harness for terminal features; use httpx/curl harness for API-only features.
Output: Story/journey tests that exercise the feature end-to-end through its actual interface. Every test references its EARS-{ID} and the Gherkin scenario it satisfies.
Coverage gate: Total line coverage must be ≥ 100% before this phase completes.
Context sentinel on completion:
SENTINEL::STORY_PROVEN::ROADMAP-{N}::PASS::{story_test_count}::{final_line_coverage_pct}::{final_branch_coverage_pct}
→ REVIEWER panel (SECURITY-REVIEWER, REGRESSION-REVIEWER, COVERAGE-REVIEWER, PERFORMANCE-REVIEWER) evaluates before Phase 6
Spawn: ds-step-7-sync, then ds-step-8-commit-message, then ds-step-9-commit-push in sequence
Prerequisite: STORY_PROVEN sentinel must be present.
Input: All accumulated sentinels + final green test run output
Output: Feature committed and pushed; roadmap STATUS updated to COMPLETE
Steps within this phase:
ds-step-7-sync): rebase against upstream; re-run tests after rebaseds-step-8-commit-message): write WHY/WHAT/TESTING/ROADMAP message per
${CLAUDE_PLUGIN_ROOT}/knowledge/protocols/commit-message.mdds-step-9-commit-push): stage, commit, push; update roadmap STATUS → COMPLETEContext sentinel on completion:
SENTINEL::DELIVERY_COMPLETE::ROADMAP-{N}::COMPLETE::{commit_hash}
→ Item is marked COMPLETE in roadmap. IDEA_COST recording (§12) is triggered.
Stack-specific agents spawned on demand by IMPLEMENT-AGENT and STORY-AGENT. Each handler carries: STACK knowledge, SUBJECT_MATTER_UNDERSTANDING, KAIZEN covenant.
Extend the pool without modifying this file — add new handlers as agent files
in ${CLAUDE_PLUGIN_ROOT}/agents/handler-{stack}.md.
| Handler | Stack | Spawned when |
|---|---|---|
| handler-architect | Pattern decision (language-agnostic) | LEAD ENGINEER §5 or IMPLEMENT-AGENT when a new boundary/context is introduced |
| PYTHON-AGENT | Python + pytest | Backend Python work |
| JS-AGENT | JavaScript/TypeScript + jest/vitest | Frontend or Node work |
| CSS-AGENT | CSS/SCSS + accessibility | Styling and layout |
| REACT-AGENT | React + React Testing Library | Component work |
| VANILLA-JS-AGENT | Vanilla JS + DOM + @testing-library/dom | Vanilla-JS / frontend design-system work |
| FASTAPI-AGENT | FastAPI + httpx + pytest | Python API work |
| PLAYWRIGHT-AGENT | Playwright (npx playwright test) | E2E story tests |
| RUST-AGENT | Rust + thiserror + proptest + clippy -D warnings | Rust libraries, CLIs, services, domain cores |
| RUST-WEBAPP-AGENT | Rust/WASM (Dioxus) + Vercel official Rust runtime | Full Rust web app + serverless API one-shot rollout (via the rust-webapp-rollout skill) |
| handler-rust-tauri | Tauri v2 desktop shell over a pure Rust core (IPC, capability ACL, webview) | A src-tauri/ crate / tauri.conf.json desktop app |
| handler-github-actions | GitHub Actions CI/CD-as-code (workflow YAML, matrix, OIDC, SHA-pinned actions) | .github/workflows/*.yml, composite/reusable actions |
| handler-roadmap-decomposition | Atomic-job breakdown of a ROADMAP item (INVEST slices, dependency graph, phase mapping) | LEAD ENGINEER §5 when a heavy roadmap item must be split into parallelisable jobs — it plans, it does not orchestrate |
When the LEAD ENGINEER identifies a stack not in the pool above, note it in
doc/FOUNDRY_PLAN.md under ## VALUE_HANDLER_POOL Required with a description
of what knowledge the new handler needs. The self-improvement protocol (§14)
will formalise it into an agent file.
A REVIEWER panel is invoked at every phase transition. Panels are composable —
each transition uses a specific subset of reviewers. All reviewers carry the
KAIZEN covenant. Spawn using reviewer agent with a role parameter.
| Transition | Reviewers invoked |
|---|---|
| → ADR (when architect spawned) | ARCHITECTURE-REVIEWER |
| → FEATURE (after EARS) | EARS-REVIEWER, SMU-REVIEWER |
| → TEST (after FEATURE) | BDD-REVIEWER, COVERAGE-REVIEWER |
| → IMPLEMENT (after TEST) | TEST-DESIGN-REVIEWER, COVERAGE-REVIEWER |
| → STORY (after IMPLEMENT) | DESIGN-REVIEWER, COVERAGE-REVIEWER, PERFORMANCE-REVIEWER |
| → DELIVERY (after STORY) | SECURITY-REVIEWER, REGRESSION-REVIEWER, COVERAGE-REVIEWER, PERFORMANCE-REVIEWER |
| Verdict | Meaning | Action |
|---|---|---|
PASS | Output accepted | Pipeline continues to next phase |
NEEDS_REVISION | Specific issues found | Feedback returned to phase agent; phase reruns |
BLOCK | Critical issue | Escalates to LEAD ENGINEER; pipeline pauses |
If a phase receives NEEDS_REVISION 3 times in sequence, escalate to BLOCK
automatically and surface the issue to the user.
See ${CLAUDE_PLUGIN_ROOT}/knowledge/protocols/context-sentinel.md for full format specification.
Sentinels are structured strings appended to agent context at each phase boundary. They accumulate — by Phase 6 every agent holds the full chain:
SENTINEL::EARS_COMPLETE::ROADMAP-7::PASS::EARS-042,EARS-043,EARS-044
SENTINEL::FEATURE_COMPLETE::ROADMAP-7::PASS::9::features/user-auth.feature
SENTINEL::TESTS_WRITTEN::ROADMAP-7::RED::27::EARS-042,EARS-043,EARS-044
SENTINEL::IMPL_COMPLETE::ROADMAP-7::GREEN::4::98.3
SENTINEL::STORY_PROVEN::ROADMAP-7::PASS::6::100.0
SENTINEL::DELIVERY_COMPLETE::ROADMAP-7::COMPLETE::a1b2c3d
This chain provides every downstream agent with a traceable record of what was produced at each stage, without requiring conversation history.
See ${CLAUDE_PLUGIN_ROOT}/knowledge/testing/test-policy.md for full detail.
100% line coverage is the floor, not the goal.
The test pyramid above that floor:
| Layer | Tool | Gate |
|---|---|---|
| Unit | pytest / jest | 100% line coverage required |
| Integration | pytest / jest + test doubles | All service interactions exercised |
| Behavioural (BDD) | pytest-bdd / cucumber / behave | Every Gherkin scenario passing |
| Story (E2E) | Playwright / Selenium | Every user journey exercised |
Every test must reference its EARS ID. Every story test must trace to a Gherkin scenario. The COVERAGE-REVIEWER enforces this at every review gate.
On STORY_PROVEN for any roadmap item, append one JSON line to IDEA_COST.jsonl
(at project root or doc/). File is append-only — never modify existing records.
See ${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/idea-cost-schema.md for the full authoritative schema with all
fields, nested structure, type definitions, and example records.
Do not use an inline schema here — any divergence from the reference creates
split-brain records that break the estimation feedback loop. When writing a
record, open ${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/idea-cost-schema.md and follow it exactly.
After each item, compute estimation_accuracy_pct:
accuracy = (1 - abs(actual_tokens - estimated_tokens) / estimated_tokens) * 100
Surface accuracy trends at tier boundaries. Over time, LEAD ENGINEER uses this data to improve estimates. See §14 (self-improvement).
If the i2p plugin is installed and the project has an active lifecycle (.i2p/lifecycle.json exists),
record this item's authoritative all-agent token total into the lifecycle's BUILD phase so the
status-line ◈ life widget and the estimate↔actual calibration reflect real build cost (not just the
main-thread transcript the Stop hook sees):
bash <i2p>/skills/lifecycle/scripts/cost.sh record . BUILD <token_accounting.tokens_total>
Resolve <i2p> by capability (the installed i2p plugin root); skip silently if i2p is absent. This is the
authoritative BUILD actual — see i2p/knowledge/instrumentation.md.
IDEA_COST.jsonl is the only place real tokens/wall-clock are captured (here, mid-cycle). At the same
STORY_PROVEN milestone, run the scorecard skill (/foundry:scorecard) to snapshot the project's
artifact-measured quality — scorecard.sh reads this record for the real cost fields and combines it with
coverage, corpus FP-rate, rule/test counts, and the SENTINEL verdict into SCORECARD.json. That snapshot,
compared run-over-run, is the proof the product is getting better. Schema:
${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/scorecard-schema.md.
Inspection is user-initiated only. Say "inspect FOUNDRY" / "run the inspector" to trigger. There is no scheduled or automatic invocation.
See agents/inspector.md for full behaviour. In summary:
${CLAUDE_PLUGIN_ROOT}) — skills, agents, knowledge,
commands, hooks, manifests — and the companion plugins (sentinel/pressroom) if present.FOUNDRY_INSPECTION_REPORT.md in the current project (never outside it) with
severity-ranked findings.Note: there is no git lock — a plugin install is read-only, not a shared mutable repo, so there is nothing to lock or push. The inspector mutates files only when run inside the marketplace's own source repository, and never commits/pushes unless the user asks.
FOUNDRY carries the KAIZEN self-improvement covenant. At the end of any full roadmap cycle, or when the daily inspector surfaces proposals:
FOUNDRY_INSPECTION_REPORT.md.commit-message.md):
git add <files>
git commit -m "skill: <description of improvement>"
The model-tier policy is canonical in exactly one place —
${CLAUDE_PLUGIN_ROOT}/knowledge/policy/model-selection.md.
It carries: which work class runs on which tier (opus / sonnet / haiku) and why; how the
model: inherit value-handlers are spawned per phase (TEST → haiku, IMPLEMENT → sonnet,
STORY → opus); and the tier → concrete-model-ID mapping (resolved at spawn time, never hardcoded).
Do not restate that table here. Reference it, so the whole fleet re-tiers in one edit and no pinned ID can silently age out. (This pointer deliberately replaces a former duplicate of the policy, whose independent ageing is exactly what produced a stale-model-ID drift across 13 sites.)
A handler asked to write test code on opus, or stories on haiku, must refuse and surface the model mismatch to the orchestrator.
Phase 5 (ds-step-story-tests) emits the STORY_PROVEN sentinel and is the
only step authorised to do so. No item may advance to Phase 6 (DELIVERY) without
STORY_PROVEN in its sentinel chain.
| Sentinel | Owner | Status when emitted |
|---|---|---|
STORY_PROVEN | ds-step-story-tests (Phase 5) | PASS (line+branch coverage = 100%, UI gate passed, performance assertions pass) |
DELIVERY_COMPLETE | ds-step-9-commit-push (Phase 6) | COMPLETE (commit pushed, roadmap closed) |
STORY_PROVEN and DELIVERY_COMPLETE are different signals. Past versions of
this pipeline conflated them by having step-9 emit STORY_PROVEN — that was a
bug. IDEA_COST.jsonl is written only when both are present in the chain.
| File | Purpose | When to read |
|---|---|---|
${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/tier-assignment.md | PRIORITY_STATUS → tier logic + budget caps | Before §4 |
${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/idea-cost-schema.md | IDEA_COST.jsonl field definitions + examples | Before §12 |
${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/subject-matter-understanding.md | SMU document template | Before §5 (LEAD ENGINEER) |
${CLAUDE_PLUGIN_ROOT}/knowledge/protocols/context-sentinel.md | Sentinel format and accumulation protocol | Before §10 |
${CLAUDE_PLUGIN_ROOT}/knowledge/testing/test-policy.md | Test pyramid + coverage standards + enforcement | Before §11 |
${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/agent-roster.md | All agent roles, capabilities, spawn conditions | Before §7–§9 |
${CLAUDE_PLUGIN_ROOT}/knowledge/architecture/kaizen-covenant.md | KAIZEN replication covenant | Carried by all artefacts |
${CLAUDE_PLUGIN_ROOT}/knowledge/protocols/definition-of-done.md | Project DoD template — copy to project root | Before §7 (per-item) |
${CLAUDE_PLUGIN_ROOT}/knowledge/orchestration/orchestration-loop.md | Per-item loop state model + stage routing table | Before §7 (per-item) |
${CLAUDE_PLUGIN_ROOT}/skills/lifecycle-states/states/discovery.md | Pre-step-0 readiness exit criteria | Before spawning step-0 |
${CLAUDE_PLUGIN_ROOT}/skills/lifecycle-states/states/specification.md | EARS and Gherkin exit criteria (steps 1–2) | Before advancing to step-3 |
${CLAUDE_PLUGIN_ROOT}/skills/lifecycle-states/states/verification.md | Test-first and green-run exit criteria (steps 3–6) | Before advancing to step-7 |
${CLAUDE_PLUGIN_ROOT}/skills/lifecycle-states/states/delivery.md | Sync, commit, and push exit criteria (steps 7–9) | Before global DoD audit |
${CLAUDE_PLUGIN_ROOT}/skills/lifecycle-states/states/production-readiness.md | Final DoD certification (post step-9) | Global DoD audit |
This document is part of the FOUNDRY skill in the idea-to-production marketplace. All artefacts produced by FOUNDRY — plans, specs, feature files, test files, cost records, and inspection reports — carry this covenant. Its spirit is kaizen — continuous improvement, never finished.
Skill: FOUNDRY — idea-to-production marketplace
When the conveyor delivers a shippable item (it reaches SHIP / the item is marked COMPLETE), and the i2p plugin is installed, mark the BUILD phase done so the marketplace product lifecycle and the status line advance to ASSURE:
/i2p-lifecycle done BUILD # order-safe & idempotent — a no-op unless a lifecycle is running at BUILD
Degrades silently when i2p is absent. The canonical model is i2p/knowledge/product-lifecycle.md.
npx claudepluginhub agentic-underground/idea-to-productionProvides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.