devrites 2.6.0 → 3.0.0

package/pack/generated/codex/skills/rite-build/reference/one-slice-cycle.md

@@ -0,0 +1,42 @@
+# One-slice cycle
+
+The discipline that makes large features manageable: build one thin slice, leave it
+working and proven, stop. Never implement the whole feature in one pass.
+
+## The cycle
+The orchestrator (`$rite-build`) gates and records; the **wright** writes. See
+[`wright-dispatch.md`](wright-dispatch.md).
+```
+SELECT    → orchestrator: restate slice goal + acceptance + scope boundary; HITL gate (pause pre-code)
+(SHAPE)   → orchestrator: if UI and no design-brief.md, shape it (devrites-ux-shape) before dispatch
+(FORGE)   → orchestrator: if slice is Forge: yes, compete K=2–3 isolated candidate wrights →
+            devrites-forge-judge → land one winner → forge-report.md, then continue at DOUBT (forge.md)
+DISPATCH  → hand the slice contract to devrites-slice-wright (fresh context). Inside the wright:
+   ORIENT    → load only the files this slice touches; learn the project's idiom; reuse-first
+   (RED)     → if behavior change: write the failing test first
+   IMPLEMENT → smallest complete version; match conventions; devrites-frontend-craft for UI
+   VERIFY    → targeted tests (+ types/lint/build); fail-on-red; return artifact — code + tests only
+(DOUBT)   → orchestrator, on return: devrites-doubt each non-trivial decision the wright stood up
+PROVE     → orchestrator: fail-on-red check on the wright's gates; browser proof for UI
+RECORD    → orchestrator: state.md, evidence.md, touched-files.md (the canonical writer)
+(CHECKPOINT) → orchestrator: if .devrites/CHECKPOINT is set, commit the proven slice
+            local-only as WIP(<slug>) with a [devrites-context] body → survives a crash (checkpoint.md)
+STOP      → report + recommend next; do not start slice N+1
+```
+
+## Why stop after one slice
+- Keeps the diff reviewable (~one capability, roughly ≤100 lines of meaningful change).
+- Surfaces integration/drift problems while they're cheap.
+- Gives the user a decision point — they may reprioritize, reslice, or polish now.
+- Prevents the "90% done" pile-up where nothing is actually proven.
+
+## Restate the scope boundary
+Before coding, write what this slice will and will **not** touch. This is the contract
+you check yourself against — anything outside it is scope creep or a drift event.
+
+## When the slice can't be completed cleanly
+- Discovered the plan is wrong → **Spec Drift Guard** (stop, record, classify, maybe
+  ask, `$rite-plan` repair).
+- Slice is bigger than one cycle → stop and `$rite-plan reslice`.
+- Blocked on a failing dependency → record in `state.md`, `$rite-plan unblock`.
+Never power through by guessing.

package/pack/generated/codex/skills/rite-build/reference/output.md

@@ -0,0 +1,41 @@
+# rite-build output
+
+This is the normative output contract for `$rite-build`. `../SKILL.md` points
+here so the root skill can stay focused without weakening response shape.
+
+Run `devrites-engine progress` first. It prints the deterministic header, slice
+meter, and flow ribbon. Put the compact fact lines below it using the shared
+completion reply contract
+([`devrites-lib/reference/reply-contract.md`](../../devrites-lib/reference/reply-contract.md)).
+
+**Slice built; slices remain:**
+```text
+Done: built slice <n> — <name>.
+Changed: <files touched>; state.md, touched-files.md, evidence.md
+Evidence: acceptance met; tests <cmd -> pass>; browser <summary | n/a>; drift <none|handled>
+Open: <none | blockers | questions>
+Next: $rite-build
+Record: .devrites/work/<slug>/evidence.md
+↻ Hygiene: /clear between slices; $rite-handoff if away > a few hours
+```
+
+**Slice built; all slices built:**
+```text
+Done: built slice <n> — <name>; all slices built.
+Changed: <files touched>; state.md, touched-files.md, evidence.md
+Evidence: acceptance met; tests <cmd -> pass>; browser <summary | n/a>; drift <none|handled>
+Open: <none | blockers | questions>
+Next: $rite-prove
+Record: .devrites/work/<slug>/evidence.md
+↻ Hygiene: /clear before $rite-prove; $rite-handoff if away > a few hours
+```
+
+For awaiting-human or stopped states, use the typed templates from the reply contract.
+Keep fact lines terse. The meter and ribbon carry progress; don't duplicate them in
+prose.
+
+For a **forged slice**, the `Built` line names the competition and points at the record, e.g.
+`Done: built slice 3 — csv-streaming (forged: 3 candidates, winner B).` and
+`Record: .devrites/work/<slug>/forge-report.md`.
+
+**DO NOT continue to the next slice automatically** — even at `✅ ALL BUILT`, `$rite-prove` is the user's call.

package/pack/generated/codex/skills/rite-build/reference/phase-contract.md

@@ -0,0 +1,258 @@
+# rite-build phase contract
+
+See also [`one-slice-cycle.md`](one-slice-cycle.md).
+
+## Execution spine
+
+Run the engine gates at these moments:
+
+```bash
+devrites-engine preamble
+devrites-engine snapshot
+devrites-engine build-readiness; echo "readiness rc=$?"
+
+devrites-engine reconcile snapshot
+devrites-engine stuck log "$(cat .devrites/ACTIVE 2>/dev/null)" dispatch "<slice id>"
+
+devrites-engine footprint log <slug> doubt "<decision id>"
+
+devrites-engine reconcile check; echo "reconcile rc=$?"
+devrites-engine test-integrity; echo "test-integrity rc=$?"
+devrites-engine package-existence; echo "package-existence rc=$?"
+
+devrites-engine conventions contradict --key <key> --slug <slug> \
+  --evidence "<what the live code does>" --drift-file .devrites/work/<slug>/drift.md
+devrites-engine footprint log <slug> wright "<slice id>"
+devrites-engine tick-afk <state.md path>
+
+devrites-engine progress
+```
+
+Conditional calls still obey the detailed rules below: doubt logging runs for
+each stood decision; convention contradiction runs only when the wright reports
+or exposes a contradiction; `tick-afk` runs only under AFK; forge follows
+[`forge.md`](forge.md) and still returns to the same doubt, record, and stop
+gates.
+
+> **Running the gate helpers.** Each gated `bash` block calls the installed engine
+> directly: `devrites-engine <command> [args]`. It **propagates its exit code**,
+> so the `rc=$?` checks below still hold; host hooks fail open when the binary is absent,
+> but phase gates should treat a missing engine as a setup problem.
+
+0. **Rules + AFK + readiness check.** Read `.agents/skills/devrites-lib/reference/standards/core.md` first. Then **run the
+   shared orientation preamble** — it prints `state.md`, the artifacts present, the run
+   mode (HITL/AFK), and the open-question tally by gate, deterministically:
+   ```bash
+   devrites-engine preamble
+   devrites-engine snapshot
+   ```
+   Treat the snapshot as the canonical machine-readable status; if you need to tell the
+   user the next command, use `nextCommands.claude` in Claude Code or `nextCommands.codex`
+   in Codex. Then **run the readiness gate** — it enforces the step-0 stop conditions by exit code,
+   not by memory:
+   ```bash
+   devrites-engine build-readiness; echo "readiness rc=$?"
+   ```
+   A non-zero `rc` is a hard STOP: `2` → `$rite-define` (plan not approved), `3` →
+   `$rite-resolve` (awaiting human), `4` → `$rite-plan` (blocked). The prose below is the
+   same gate for installs without the script.
+   Orient from its digest. If `Status == awaiting_human` → **STOP**, tell the user to run
+   `$rite-resolve <qid> "<answer>"`. If `state.md` has no `Plan approved: <iso>` field
+   → **STOP**, tell the user the plan isn't approved yet (`$rite-define` writes it when
+   the human confirms). If `.devrites/AFK` is present, re-derive the remaining AFK budget
+   from `state.md`'s `AFK slices remaining: <n>` field (initialized from `.devrites/AFK`
+   `max_slices` on the first AFK build); if it is `0` → **STOP** (forced HITL stop; raise
+   the count in `state.md` or remove the sentinel to continue). See
+   [`afk-discipline.md`](afk-discipline.md).
+1. Read `spec.md`, `plan.md`, `tasks.md`, `assumptions.md`, `drift.md`, and `test-plan.md`
+   if present (the vetted coverage target from `$rite-vet` — the slice's tests come from
+   here when it exists). `state.md` and the open-`questions.md` tally are already in the
+   preamble digest from step 0 — re-read `questions.md` only for the full text of a flagged
+   blocking question.
+   If a **blocking `[NEEDS CLARIFICATION]`** remains or the spec/plan readiness gates
+   don't pass, stop → `$rite-spec` (to resolve) or `$rite-plan` (to repair). Don't build
+   on an unresolved spec.
+2. Select the next pending slice (or the one in `$ARGUMENTS`). **Restate its goal,
+   acceptance criteria, and scope boundary** in one short block. Confirm it's still the
+   right next slice. Write the slice's `Mode` to `state.md` as `Slice mode: <HITL|AFK>` on
+   **every** selection (not only on the HITL pause path); `$rite-resolve` clears or updates
+   it on resume.
+2a. **HITL gate (pre-action pause).** Read the slice's `Mode`. If `HITL`, surface the
+    checkpoint as a ranked **option set** and resolve it **before** any code lands, per
+    [`checkpoint-protocol.md`](checkpoint-protocol.md). Branch on whether
+    a human is here:
+    - **Human present (interactive — no `.devrites/AFK`) → ask inline via `AskUserQuestion`.**
+      This is the default for an interactive build. Render 2-4 options, the recommended one
+      **first and labelled `(Recommended)`**, each with its dimension-tagged rationale + the
+      trade-off it accepts, plus the `Something else — I'll describe it` escape hatch. The
+      human picks; record the pick to `questions.md` (`answered`) + `decisions.md` (through the
+      `devrites-engine resolve` writer), clear the gate, and **continue to step 3 in place** — do **not**
+      STOP, do **not** route through `$rite-resolve`.
+    - **Human absent / AFK (`.devrites/AFK` present) → auto-pick or persist + STOP.** For a gate
+      in `allow_gates`, auto-pick the recommended option (option 1) and proceed; otherwise
+      append the `questions.md` entry, write the `Awaiting human` block to `state.md`, set
+      `Status: awaiting_human`, fire the `notify:` hook, then **STOP** — resume when the user
+      runs `$rite-resolve <qid> "<answer>"`. `blocking` / `escalating` / irreversible-risk gates
+      always take this stop path, never the AFK auto-pick.
+3. **Snapshot the tree, then dispatch the build core to `devrites-slice-wright`** — one `Task`
+   call, fresh context. **First**, capture the pre-dispatch tree so the reconcile gate (step 6)
+   can prove you never touched source — run this immediately before the `Task` call:
+   ```bash
+   devrites-engine reconcile snapshot
+   ```
+   Then log the dispatch so the stuck-loop detector can catch a slice that keeps being
+   re-dispatched without progress (it pauses the build even under AFK):
+   ```bash
+   devrites-engine stuck log "$(cat .devrites/ACTIVE 2>/dev/null)" dispatch "<slice id>"
+   ```
+   **Forge branch — only if the selected slice is `Forge: yes`.** Instead of the single dispatch
+   below, run the competitive build per [`forge.md`](forge.md): K=2-3 candidate
+   wrights, each in an **isolated git worktree** on the distinct strategy `$rite-vet` named, then a
+   fresh-context [`devrites-forge-judge`](.codex/agents/devrites-forge-judge.toml) scores them against
+   acceptance + `test-plan.md` + `.devrites/principles.md` + the anti-slop charter, and you land
+   exactly **one** winner's diff in the working tree, graft any cheap runner-up improvement by
+   continuing the winning wright once, and write `forge-report.md`. Forge returns the **same shape**
+   a single wright does (one structured artifact, for the winner), so steps 4-7 (doubt, fail-on-red,
+   reconcile against the winner's claimed set, record, stop) run **unchanged**. If you cannot give
+   the judge an objective scorecard (the slice lacks acceptance / `test-plan.md` coverage) or cannot
+   name two genuinely different strategies, the flag is stale — clear it and build single-path. The
+   single-writer invariant is intact: each candidate owns its own tree, exactly one author's diff
+   lands. Then **stop** here for this slice; the steps below are the default single-path dispatch.
+
+   Then assemble the slice contract and send it per
+   [`wright-dispatch.md`](wright-dispatch.md): the slice goal, acceptance
+   criteria, and **scope boundary**; the paths it may touch (`touched-files.md`); the context
+   paths to read (`spec.md`, `plan.md`, `decisions.md`, `assumptions.md`, `.devrites/principles.md`
+   when present — the binding invariants the slice must honor — plus `test-plan.md`
+   when present — its per-gap test requirements + regression-criticals for this slice are the
+   coverage the wright must write — and `design-brief.md`
+   when the slice touches UI per [`frontend-trigger.md`](frontend-trigger.md)); and the
+   `.agents/skills/devrites-lib/reference/standards/` files in scope. The wright **orients** on the project's idiom (using a
+   code-intelligence index **if available** — `codebase-memory-mcp` first, cross-checked with
+   `codegraph` (`.codegraph/` / `codegraph_*`) + `graphify` (`graphify-out/`), else standard
+   methods (LSP / `Read`/`Grep`/`Glob`); see `.agents/skills/devrites-lib/reference/standards/tooling.md` — for
+   placement/callers/impact), writes the **failing test first** when
+   behaviour changes ([`tdd.md`](tdd.md)), implements the **smallest complete** version in
+   the project's style (applying `devrites-frontend-craft` to `design-brief.md` for UI, and
+   `devrites-source-driven` — with context7 if available — for uncertain framework facts), runs the slice's **targeted tests**
+   (plus typecheck / lint / build where the project has them), and returns a structured artifact
+   — **code + tests only; it does not write the workspace files.** If the slice is UI but no `design-brief.md` exists (e.g. a spec written before
+   shaping), shape it via `devrites-ux-shape` before the wright codes. If the `Task` tool is
+   unavailable, run the wright's discipline **inline** as a flagged fallback (see the reference)
+   — same one-slice cycle, no isolation; in that case write
+   `.devrites/work/<slug>/.reconcile-inline` so the reconcile gate skips (you are legitimately
+   the writer in this fallback).
+4. **Doubt the decisions it stood up.** For each entry in the wright's `Decisions stood`
+   (branching, boundary crossing, data model, auth, public API, migration, user-flow change,
+   "this is safe/scales") apply `devrites-doubt` **before accepting the slice** — the writer
+   doesn't grade its own decisions. Each `devrites-doubt` invocation **dispatches the
+   `devrites-doubt-reviewer` subagent** (doing the adversarial pass inline is the writer grading
+   itself — the thing this step exists to forbid). **Completion criterion (checkable):** step 4
+   is done only when **every** `Decisions stood` entry carries a recorded `devrites-doubt`
+   verdict — `accept`, or `reject` + the required changes — in `decisions.md` (accepted
+   trade-offs) / `questions.md` (open gates). A `Decisions stood` entry with **no verdict on
+   record** means doubt did not run for it: **do not enter step 5 and do not mark the slice
+   `built`.** Log each dispatch so the seal can prove doubt ran (the footprint already counts a
+   `doubt` kind):
+   ```bash
+   devrites-engine footprint log <slug> doubt "<decision id>"
+   ```
+   The doubt loop honours `.devrites/AFK` (see its AFK exception): findings below the slice's gate ceiling become advisory entries in `questions.md`;
+   destructive / auth / public-API concerns always pause regardless. A non-empty `Escalation` in
+   the artifact is handled here too: irreversible-risk / blockers → blocking question + set
+   `Status: awaiting_human`; a scope-changing answer → `$rite-plan repair` (Spec Drift Guard),
+   never silently into the slice. **If an irreversible-risk item shows up under the wright's
+   `Decisions stood` rather than `Escalation`**, treat that misclassification as itself a
+   blocking protocol violation — pause and re-dispatch with the item flagged out-of-bounds, do
+   **not** doubt-and-accept it. (The wright's return is the not-yet-load-bearing moment — the
+   slice isn't `built` or merged yet — so this post-return doubt is still pre-commit.)
+   **Principle check (same standing):** a wright return that breaks a declared principle
+   (reported in its `Principles` field, or that you detect against `.devrites/principles.md`) is
+   handled here like an irreversible-risk item — block, route to a human-approved scoped
+   exception in the register or stop; never doubt-and-accept a principle violation into the slice.
+5. **Fail-on-red.** If the wright's `Gates` were red (targeted tests / types / lint) or it
+   couldn't verify: do **not** mark the slice `built`, and **do not fix the code yourself**.
+   First remedy — **continue the same wright once** (`SendMessage` to it, carrying the failing
+   gate + its real output) so it fixes in its own context. This retry is for **objective
+   failures only** — red gate / type / lint / missing test coverage / UI browser-proof fail —
+   never a contested decision (that routes to `$rite-plan repair`). **Still red after the one
+   retry** → escalate: AFK → append a blocking question to `questions.md` (gate=blocking,
+   slice's SLA) + set `Status: awaiting_human`; HITL → pause as a blocking gate. Either way,
+   `Next step: $rite-plan unblock` until resolved.
+6. **Record — you are the canonical writer.** **First, run the reconcile gate (A1):** write the
+   wright's reported `Files changed` paths (one per line) to
+   `.devrites/work/<slug>/.reconcile-claimed` (or the active legacy
+   `.devrites/features/<slug>/` workspace during migration), then:
+   ```bash
+   devrites-engine reconcile check; echo "reconcile rc=$?"
+   ```
+   **Exit 5 → hard STOP:** a source file changed outside the wright's claimed set — code was
+   edited by something other than the wright (A1 breach). Revert it and re-dispatch the wright;
+   do **not** mark the slice `built`.
+
+   **Then run the test-integrity gate (anti-reward-hacking)** — prove the slice didn't reach
+   green by weakening its tests:
+   ```bash
+   devrites-engine test-integrity; echo "test-integrity rc=$?"
+   ```
+   **Exit 3 → hard STOP:** a test was deleted, skipped, or de-asserted since the slice base — the
+   slice went green by weakening its tests, a Critical protocol violation. Revert the weakening and
+   re-dispatch the wright; do **not** mark the slice `built`.
+
+   **Then run the package-existence gate (anti-hallucination)** — every new third-party import must
+   be declared in a project manifest, not just imported:
+   ```bash
+   devrites-engine package-existence; echo "package-existence rc=$?"
+   ```
+   **Exit 3 → STOP:** an imported package is not declared in any manifest (`package.json`, `go.mod`,
+   `requirements.txt`, `pyproject.toml`, `Pipfile`, `Cargo.toml`) — the classic shape of a
+   hallucinated or typo-squatted dependency. Confirm the name on the registry and declare it via the
+   package manager, or remove the import; do **not** mark the slice `built`. The gate is deterministic
+   and fail-open (not a git repo / no manifest / stdlib-only import → rc 0).
+
+   Then, from the wright's artifact, update `state.md`,
+   `evidence.md`, `touched-files.md` (and `browser-evidence.md` for UI). Add a `## Review trail`
+   to `touched-files.md`: group the slice's important `path:line` stops by concern (design intent,
+   not file order), 1–5 concerns, each stop under 15 words. This is for a later human walkthrough;
+   keep it factual and skip invented rationale. **Persist every
+   `Decisions stood` entry from the wright's artifact to a `## Decisions stood` section in
+   `decisions.md`, one line each ending `— doubt: <accept | reject-resolved | MISSING>`** — the
+   verdict from step 4, or `MISSING` if step 4 was skipped for that entry. Write this ledger at
+   record-time **independent of the doubt step**, so a stood decision lands on disk even when it
+   was never doubted: `$rite-seal` cross-checks the ledger, and a `doubt: MISSING` entry is the
+   skipped-doubt finding that would otherwise leave no trace (the per-slice skip the footprint
+   count alone can't catch). If the wright stood no decisions, write `## Decisions stood` with
+   `- none` so the absence is recorded, not assumed. If the wright reported
+   an approach it tried and backed out of, record it under a `## Dead ends` section in
+   `decisions.md` so a retry or the next agent doesn't repeat it.
+   **If the wright's `Conventions` field reports a contradiction** (the live code now
+   disagrees with a held convention), record the drift — you own this bookkeeping:
+   ```bash
+   devrites-engine conventions contradict --key <key> --slug <slug> \
+     --evidence "<what the live code does>" --drift-file .devrites/work/<slug>/drift.md
+   ```
+   It lowers the convention's band (or retires it) and appends a `convention-drift` entry to
+   `drift.md`. The ledger is **promoted only at `$rite-seal` on GO** — build only records
+   contradictions, never new conventions. **Evidence is the
+   wright's real command output, not its say-so.** Capture per
+   [`evidence-standard.md`](evidence-standard.md). Append a footprint record for this
+   slice's wright dispatch (deterministic run-weight bookkeeping the seal reports):
+   ```bash
+   devrites-engine footprint log <slug> wright "<slice id>"
+   ```
+   If `.devrites/AFK` is present, decrement
+   the budget by running `devrites-engine tick-afk <state.md path>` —
+   it decrements `state.md`'s `AFK slices remaining` field, prints the new value, and exits `3`
+   when it hits 0. **Exit 3 → STOP** (forced HITL stop; the cap is exhausted). Never rewrite
+   `.devrites/AFK` `max_slices` in place — it is read-only initial budget.
+7. **STOP.** Render the progress footer, then report and recommend the next step. Run the
+   shared footer (mirror of the step-0 preamble) so the slice meter + flow ribbon are
+   deterministic, not model-typed:
+   ```bash
+   devrites-engine progress
+   ```
+   It reads `state.md` (already updated in step 6), so the meter reflects the slice you
+   just built. **When every slice is built** (`✅ ALL BUILT`) say so explicitly — the build
+   phase is complete, the next phase is `$rite-prove` — don't leave completion implicit.
+
+> **Mid-flight discipline.** The wright (or you, in the inline fallback) must resist doing two slices, skipping TDD, adding a defensive check, or wandering outside `touched-files.md`; you must resist skipping the post-return `devrites-doubt` because the wright "seems confident" — see [`anti-patterns.md`](anti-patterns.md). Load it the moment you reach for the excuse.

package/pack/generated/codex/skills/rite-build/reference/spec-drift-guard.md

@@ -0,0 +1,43 @@
+# Spec Drift Guard (build phase)
+
+Active throughout `$rite-build`. The spec is living, not sacred — but you may not
+silently code against a plan you know is wrong. This is the canonical Spec Drift Guard;
+other phases reference it here.
+
+## Drift has occurred when
+- the spec says X but the code makes X impossible/wrong;
+- the plan assumes a file/API/component/route/model/command that doesn't exist;
+- tests show the acceptance criteria are incomplete or wrong;
+- browser evidence shows the intended UX doesn't work as described;
+- an official doc contradicts the planned approach;
+- the design system / existing UX pattern contradicts the planned UI;
+- a missing user decision that affects product behavior surfaces.
+
+## Workflow
+```
+1. STOP coding.
+2. Record in drift.md (assumed vs observed, when).
+3. Classify: requirement ambiguity | implementation-plan error | codebase reality
+   mismatch | design-system mismatch | test/evidence mismatch | external-doc mismatch
+   | user-decision required.
+4. Local repair that preserves behavior/scope/architecture/data/UX/security/migration?
+     YES → log in drift.md + decisions.md, $rite-plan repair, resume.
+     NO  → ask the user (format below). Don't continue on the old plan.
+5. Run $rite-plan repair before resuming.
+```
+
+## User question format
+```
+I hit spec drift:
+- Spec/plan assumed: ...
+- Code/evidence shows: ...
+- Why it matters: ...
+
+Which direction should DevRites take?
+1. Keep the requirement, change architecture by ...
+2. Adjust the requirement to match existing behavior ...
+3. Split into a follow-up feature ...
+4. Custom: describe the intended behavior
+```
+
+Never continue coding on a known-wrong plan. Always re-plan before resuming.

package/pack/generated/codex/skills/rite-build/reference/tdd.md

@@ -0,0 +1,26 @@
+# TDD / Prove-It pattern
+
+When a slice changes behavior, prove the change with a test that fails first — so you
+know the test can fail, and that your code is what makes it pass.
+
+## Red → Green → Refactor
+1. **Red** — write the smallest test that expresses the new behavior. Run it; watch it
+   fail for the *expected* reason (not a typo/import error).
+2. **Green** — write the minimum code to pass. Resist gold-plating.
+3. **Refactor** — clean up with the test as a safety net. Re-run; stay green.
+
+## Prove-It (when strict TDD doesn't fit)
+For code where a test-first flow is awkward (exploratory glue, config, some UI), still
+produce evidence:
+- Capture the current (wrong/absent) behavior — a failing assertion, a screenshot, a
+  log line.
+- Make the change.
+- Capture the new behavior the same way. The before/after pair *is* the proof.
+
+## Rules
+- Use the project's existing test framework and commands (discovered in `spec.md` →
+  "Commands discovered"). Don't introduce a new test runner to prove a slice.
+- One behavior per test; name it for the behavior, not the function.
+- A green test you never saw fail proves nothing — always see red first.
+- Don't delete or weaken a failing test to "pass". If it's wrong, that may be drift.
+- Record the test command + result in `evidence.md`.

package/pack/generated/codex/skills/rite-build/reference/wright-dispatch.md

@@ -0,0 +1,140 @@
+# Slice-wright dispatch
+
+How `$rite-build` hands the **build core** of one slice to `devrites-slice-wright` — the
+fresh-context, **write-capable** executor under `.codex/agents/`. Loaded on demand by
+`$rite-build`; not a skill itself. Sibling of
+[`../../devrites-lib/reference/parallel-dispatch.md`](../../devrites-lib/reference/parallel-dispatch.md)
+(the read-only reviewer fan-out) — same isolation principle, opposite direction: this one
+**writes**.
+
+Pattern: the orchestrator owns the **gates and the workspace**; the wright owns the **writing**.
+Brief it precisely, in a clean context, with only the slice contract — then doubt, record, and
+gate its return.
+
+## Why a fresh context writes the slice
+The orchestrator's window is full of spec investigation, planning, prior slices, and tool
+output — the exact "lost-in-the-middle" load that degrades instruction-following and pulls the
+model toward generic code. The wright starts clean and sees only the contract, so it holds the
+slice boundary strictly and writes to the *project's* idiom instead of drifting. Single-threaded
+by design: **one** wright per slice, never a parallel fan-out of writers *sharing a tree* —
+concurrent writers on one working tree make conflicting implicit decisions and produce incoherent
+code. The one sanctioned exception is a **forge** slice, which competes K candidates in *isolated*
+worktrees and lands exactly one — no tree ever has two authors, so the invariant holds (see
+[Forge](#forge--competing-candidates-the-deliberate-exception)).
+
+## The contract `$rite-build` sends
+One `Task` call to `devrites-slice-wright` carrying everything the writer needs and nothing it
+doesn't:
+
+```
+Build one slice of the active DevRites feature. You have a clean context — this contract is
+the whole job.
+
+Workspace: .devrites/work/<slug>/
+Slice: <id — name>
+Goal: <one or two sentences>
+Acceptance criteria: <the slice's criteria, verbatim>
+UI visual acceptance: <if UI — state × viewport × input + target R-id/brief rule, verbatim from tasks.md>
+Scope boundary: <what it WILL and will NOT touch>
+Mode: <HITL | AFK>   (+ AFK budget note if a cap is set)
+
+Targets (stay inside these — from touched-files.md): <paths>
+Interfaces / signatures to match: <if any>
+Read yourself: spec.md, plan.md, decisions.md, assumptions.md, rite-polish/reference/anti-ai-slop.md<, design-brief.md if UI>
+Rules in scope (.agents/skills/devrites-lib/reference/standards/): coding-style, error-handling, testing, patterns<, security if input/auth/data><, performance if hot path / query / large payload>
+Test completeness: write ≥1 asserting test for EVERY interactive element + user flow in this
+  slice's test-plan.md interaction inventory, each at the right level (fields/elements →
+  unit/component; critical journeys → one E2E; never one-per-field). No element ships
+  unverified — testing.md "Completeness". If the slice has no test-plan inventory, derive the
+  element/flow list from the slice's own UI surface and cover it the same way.
+
+Apply your documented discipline (orient → RED → implement smallest complete → verify →
+return). Frontend slice → build to design-brief.md with devrites-frontend-craft and close
+the slice's visual-acceptance deltas before returning. Uncertain
+framework fact → verify at the source. Code + tests only — do NOT write the workspace
+bookkeeping files; return that data. Return your structured artifact, not your transcript.
+```
+
+Rules:
+- **One `Task` call, one wright.** Never dispatch two writers on the same slice.
+- **No author reasoning beyond the contract.** Give the slice spec, not your analysis of *how*
+  to code it — a clean, undirected read is the point.
+- **Name the boundary explicitly.** The scope boundary is the single most load-bearing field; an
+  underspecified one is the main cause of drift.
+
+## On return — the orchestrator's job (don't delegate these)
+**You never edit source here.** The wright is the only writer of code + tests; you write only
+`.devrites/` bookkeeping. Every remedy below is **continue the same wright once** or **stop +
+escalate** — never an inline patch. You snapshotted the tree before dispatch (`devrites-engine reconcile snapshot`); the reconcile check in step 4 proves no source changed outside the wright's claimed
+set.
+
+1. **Doubt the surfaced decisions.** For each entry in the wright's `Decisions stood`, apply
+   `devrites-doubt` (→ `devrites-doubt-reviewer`) before accepting — the writer must not grade
+   its own decisions. The wright's return is the not-yet-load-bearing moment (slice not `built`,
+   not merged), so this post-return doubt is still pre-commit. Irreversible-risk items always
+   pause — and an irreversible item that the wright filed under `Decisions stood` instead of
+   `Escalation` is a protocol violation: pause and re-dispatch with it flagged out-of-bounds,
+   don't doubt-and-accept.
+2. **Honor escalations.** A non-empty `Escalation` → do **not** mark the slice built. Write the
+   `questions.md` entry + `state.md` `Awaiting human` (blocking gate), or route a scope change
+   through `$rite-plan repair` (Spec Drift Guard). You are the canonical writer of these files.
+3. **Fail-on-red.** If `Gates` show red (or the wright couldn't verify), the slice is **not
+   built** — and you do **not** fix the code. First remedy: **continue the same wright once**
+   (`SendMessage`, carrying the failing gate + real output) so it fixes in its own context —
+   objective failures only (red gate / type / lint / missing coverage / UI browser-proof fail),
+   never a contested decision. Still red after that one retry → blocking question (AFK) or
+   blocking gate (HITL); `Next: $rite-plan unblock`.
+   An interactive element or user flow in the slice's test-plan interaction inventory left with
+   **no asserting test** has the same standing as red: an unverified-element gap blocks the
+   slice (don't mark it built). Continue the same wright to cover it, or record a blocker.
+4. **Reconcile, then record.** First prove A1 held: write the wright's `Files changed` paths
+   (one per line) to `.devrites/work/<slug>/.reconcile-claimed` and run `devrites-engine reconcile check`.
+   **Exit 5 → STOP** — a source file changed outside the wright's claimed set (A1 breach); revert
+   it and re-dispatch, don't mark the slice built. Then persist the wright's artifact to
+   `state.md`, `evidence.md`, `touched-files.md` (and `browser-evidence.md` for UI) per
+   [`evidence-standard.md`](evidence-standard.md). Evidence is the wright's real command output,
+   not its say-so. Add a concern-ordered `## Review trail` to `touched-files.md` from the wright's changed paths and summary so a human can review by design intent instead of file order. **Persist every `Decisions stood` entry to a `## Decisions stood` section in
+   `decisions.md`, one line each ending `— doubt: <accept | reject-resolved | MISSING>`** —
+   independent of the doubt step (step 1 above), so a skipped decision still lands on record for
+   the seal's doubt-coverage cross-check (`- none` when the wright stood nothing). Then tick AFK if
+   `.devrites/AFK` is present (`devrites-engine tick-afk`; exit 3 → STOP).
+
+## Forge — competing candidates (the deliberate exception)
+
+The single-writer rule forbids parallel writers **sharing one tree**. A `Forge: yes` slice
+(flagged by `$rite-vet` as a genuine architecture fork at Complexity ≥4) is the one sanctioned
+fan-out, and it keeps the rule intact by **isolation**: each candidate wright works in its own
+`git worktree`, sees the identical slice contract plus one **distinct strategy**, and never
+touches another candidate's tree. A read-only [`devrites-forge-judge`](.codex/agents/devrites-forge-judge.toml)
+then scores the finished candidates against acceptance + `test-plan.md` + `.devrites/principles.md`
++ the anti-slop charter, and the orchestrator lands **exactly one** winner's diff in the working
+tree. No tree ever has two authors; exactly one author's work ships. Everything downstream (doubt,
+fail-on-red, reconcile against the winner's claimed set, record) runs on the winner as if a single
+wright had built it.
+
+Full mechanics — strategy derivation, worktree setup, the judge contract, landing + grafting the
+winner, `forge-report.md`, AFK budgeting, and the worktree-unavailable fallback — live in
+[`forge.md`](forge.md).
+
+## Fallback
+If the `Task` tool / sub-agent dispatch is unavailable, `$rite-build` runs the wright's
+discipline **inline** in its own context and flags it as a fallback (no clean-context benefit).
+The slice still gets the full one-slice cycle — orient → RED → implement → verify — under the
+same anti-slop charter; it just doesn't get the isolation. In this path the orchestrator is
+legitimately the writer, so write `.devrites/work/<slug>/.reconcile-inline` before editing — the
+reconcile gate (step 4) skips when that sentinel is present. Mirrors the reviewer-dispatch
+fallback in
+[`../../devrites-lib/reference/parallel-dispatch.md`](../../devrites-lib/reference/parallel-dispatch.md).
+
+## Optional pre-block hook (defense in depth)
+`devrites-engine reconcile` is the **post-hoc** gate — it always runs and catches an A1 breach at record time.
+A companion **pre-block** hook, `devrites-engine hook a1-guard` (a `PreToolUse` matcher on
+`Edit|Write|MultiEdit`), stops the breach *before* the write lands. It is armed only inside the
+mid-build window (between `devrites-engine reconcile snapshot` and a clean `check`, keyed on `.reconcile-base`),
+allows the wright (subagent calls carry `agent_id`), the inline fallback (`.reconcile-inline`),
+and any `.devrites/` write — so it never touches `$rite-polish`, `$rite-quick`, or ordinary
+manual edits. It ships **observe-only** (logs would-be blocks to `.a1-guard.log`, never blocks);
+flip to enforce with `DEVRITES_A1_HOOK=enforce` or a `.devrites/work/<slug>/.a1-enforce` file once
+the log confirms it never flags the wright's own edits (older Claude Code builds may not populate
+`agent_id` — the log is the proof before you enforce). The post-hoc gate stands on its own; the
+hook is belt-and-suspenders.