devrites 7.3.3 → 7.3.5

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (72) hide show
  1. package/CHANGELOG.md +44 -3
  2. package/README.md +21 -1
  3. package/docs/adr/0039-autocomplete-structural-admission.md +87 -0
  4. package/docs/adr/README.md +1 -0
  5. package/docs/cli.md +182 -7
  6. package/docs/command-map.md +2 -2
  7. package/docs/engine/commands.md +53 -3
  8. package/engine/commands.go +1 -1
  9. package/engine/help.go +3 -1
  10. package/engine/internal/lib/claims.go +65 -8
  11. package/engine/internal/lib/continuation.go +38 -4
  12. package/engine/internal/lib/dispatch.go +124 -63
  13. package/engine/internal/lib/frontier.go +864 -0
  14. package/engine/internal/lib/frontier_controller.go +652 -0
  15. package/engine/internal/lib/frontier_roadmap.go +168 -0
  16. package/engine/internal/lib/handoff.go +2 -5
  17. package/engine/internal/lib/next.go +19 -0
  18. package/engine/internal/parallel/cli.go +16 -5
  19. package/engine/internal/parallel/ops.go +7 -5
  20. package/engine/internal/parallel/repository.go +23 -0
  21. package/engine/internal/release/release.go +96 -5
  22. package/engine/main.go +4 -0
  23. package/pack/.claude/skills/devrites-lib/reference/parallel-workspaces.md +106 -114
  24. package/pack/.claude/skills/devrites-lib/reference/standards/afk-hitl.md +18 -17
  25. package/pack/.claude/skills/devrites-lib/reference/workspace-artifact-schema.md +41 -38
  26. package/pack/.claude/skills/rite-autocomplete/SKILL.md +33 -38
  27. package/pack/.claude/skills/rite-autocomplete/reference/loop.md +46 -52
  28. package/pack/.claude/skills/rite-autocomplete/reference/stop-conditions.md +28 -28
  29. package/pack/.claude/skills/rite-build/reference/parallel-batch.md +19 -17
  30. package/pack/generated/claude/skills/devrites-lib/reference/parallel-workspaces.md +106 -114
  31. package/pack/generated/claude/skills/devrites-lib/reference/standards/afk-hitl.md +18 -17
  32. package/pack/generated/claude/skills/devrites-lib/reference/workspace-artifact-schema.md +41 -38
  33. package/pack/generated/claude/skills/rite-autocomplete/SKILL.md +33 -38
  34. package/pack/generated/claude/skills/rite-autocomplete/reference/loop.md +46 -52
  35. package/pack/generated/claude/skills/rite-autocomplete/reference/stop-conditions.md +28 -28
  36. package/pack/generated/claude/skills/rite-build/reference/parallel-batch.md +19 -17
  37. package/pack/generated/codex/skills/devrites-lib/reference/parallel-workspaces.md +106 -114
  38. package/pack/generated/codex/skills/devrites-lib/reference/standards/afk-hitl.md +18 -17
  39. package/pack/generated/codex/skills/devrites-lib/reference/workspace-artifact-schema.md +41 -38
  40. package/pack/generated/codex/skills/rite-autocomplete/SKILL.md +33 -38
  41. package/pack/generated/codex/skills/rite-autocomplete/reference/loop.md +46 -52
  42. package/pack/generated/codex/skills/rite-autocomplete/reference/stop-conditions.md +28 -28
  43. package/pack/generated/codex/skills/rite-build/reference/parallel-batch.md +19 -17
  44. package/pack/generated/devin/skills/devrites-lib/reference/parallel-workspaces.md +106 -114
  45. package/pack/generated/devin/skills/devrites-lib/reference/standards/afk-hitl.md +18 -17
  46. package/pack/generated/devin/skills/devrites-lib/reference/workspace-artifact-schema.md +41 -38
  47. package/pack/generated/devin/skills/rite-autocomplete/SKILL.md +33 -38
  48. package/pack/generated/devin/skills/rite-autocomplete/reference/loop.md +46 -52
  49. package/pack/generated/devin/skills/rite-autocomplete/reference/stop-conditions.md +28 -28
  50. package/pack/generated/devin/skills/rite-build/reference/parallel-batch.md +19 -17
  51. package/pack/generated/dsh/skills/devrites-lib/reference/parallel-workspaces.md +106 -114
  52. package/pack/generated/dsh/skills/devrites-lib/reference/standards/afk-hitl.md +18 -17
  53. package/pack/generated/dsh/skills/devrites-lib/reference/workspace-artifact-schema.md +41 -38
  54. package/pack/generated/dsh/skills/rite-autocomplete/SKILL.md +33 -38
  55. package/pack/generated/dsh/skills/rite-autocomplete/reference/loop.md +46 -52
  56. package/pack/generated/dsh/skills/rite-autocomplete/reference/stop-conditions.md +28 -28
  57. package/pack/generated/dsh/skills/rite-build/reference/parallel-batch.md +19 -17
  58. package/pack/generated/omp/skills/devrites-lib/reference/parallel-workspaces.md +106 -114
  59. package/pack/generated/omp/skills/devrites-lib/reference/standards/afk-hitl.md +18 -17
  60. package/pack/generated/omp/skills/devrites-lib/reference/workspace-artifact-schema.md +41 -38
  61. package/pack/generated/omp/skills/rite-autocomplete/SKILL.md +33 -38
  62. package/pack/generated/omp/skills/rite-autocomplete/reference/loop.md +46 -52
  63. package/pack/generated/omp/skills/rite-autocomplete/reference/stop-conditions.md +28 -28
  64. package/pack/generated/omp/skills/rite-build/reference/parallel-batch.md +19 -17
  65. package/pack/generated/pi/skills/devrites-lib/reference/parallel-workspaces.md +106 -114
  66. package/pack/generated/pi/skills/devrites-lib/reference/standards/afk-hitl.md +18 -17
  67. package/pack/generated/pi/skills/devrites-lib/reference/workspace-artifact-schema.md +41 -38
  68. package/pack/generated/pi/skills/rite-autocomplete/SKILL.md +33 -38
  69. package/pack/generated/pi/skills/rite-autocomplete/reference/loop.md +46 -52
  70. package/pack/generated/pi/skills/rite-autocomplete/reference/stop-conditions.md +28 -28
  71. package/pack/generated/pi/skills/rite-build/reference/parallel-batch.md +19 -17
  72. package/package.json +1 -1
@@ -12,6 +12,7 @@ Autocomplete chains existing `/rite-*` workflows without routine-phase pauses. A
12
12
  ## Arm AFK once
13
13
 
14
14
  ```yaml
15
+ mode: autocomplete
15
16
  allow_gates: [advisory, validating]
16
17
  # max_slices / max_agents / max_minutes / max_review_queue omitted: unlimited
17
18
  # continue_sequence: false # opt out; default chains after Seal GO
@@ -22,18 +23,15 @@ allow_gates: [advisory, validating]
22
23
  # notify: "<cmd>"
23
24
  ```
24
25
 
25
- Read existing sentinel first. Preserve valid bytes, except
26
- when this invocation contains `--parallel N`: then write or replace only
27
- `max_parallel: N` and leave every other key unchanged. Stop if the sentinel is
28
- malformed. Do not stop because it names `max_slices` / `max_agents` /
29
- `max_minutes` / `max_review_queue` or only `[advisory]`. If absent, write it
30
- once after clarity, and include `max_parallel: N` when that flag is present.
31
- Without the flag, seed a newly created sentinel from retained global lease capacity
32
- when present; otherwise use fresh-run default 10. Never replace a valid existing
33
- sentinel's capacity from lease/default. Never write `max_slices` from `--max-slices`.
34
- It is otherwise read-only: never rewrite it after Vet or reset it on resume. Leftover `expires_at` is ignored
35
- and never rewritten. **Failing case:** leftover `max_parallel: 1` keeps the run
36
- serial after `--parallel 5`.
26
+ Read existing AFK first. On each explicit Autocomplete invocation, upsert only
27
+ `mode: autocomplete` and, if supplied, `max_parallel: N`; preserve other fields.
28
+ Missing mode retains standalone AFK behavior until that explicit entry. Invalid
29
+ or duplicate mode fails closed; mode grants neither phase nor action authority.
30
+ If absent, create AFK after clarity with the mode above; capacity is explicit N,
31
+ else retained global lease capacity, else 10. Preserve existing capacity without
32
+ the flag. Nested phases/wakes leave AFK unchanged: no `max_slices` flag seeding,
33
+ expiry rewrite or reset. Legacy slice/agent/time/review limits do not stop this
34
+ controller. **Failing case:** `--parallel 5` retains `max_parallel: 1`.
37
35
 
38
36
  ### Derive the mutable post-vet budget
39
37
 
@@ -43,15 +41,15 @@ remaining. AFK configuration itself stays unchanged.
43
41
 
44
42
  ### Admit each unattended cycle
45
43
 
46
- Before each phase, review fan-out, recovery, or writer dispatch: cheapest
47
- readiness check first; reject overlap. Do not stop on `max_agents`,
48
- `max_minutes`, or `max_review_queue`. Count every leaf result. New activations reset activation-local counters, retaining durable slice/recovery state.
49
-
50
- [Build return](../../rite-build/reference/output.md#caller-owned-continuation).
51
- Pass `--parallel N`; ignore leftover cap/headroom, remaining-0,
52
- and `--max-slices` stops. Agent-invented dispatch, wait, tool-call, or turn budgets are not stop conditions.
53
- An explicit human request to pause or run only a bounded subset still wins.
54
- Path collisions serialize work; they do not end the run.
44
+ Before each phase, fan-out, recovery or writer: run cheap readiness and reject
45
+ overlap. Before Build admission, execute the shared immediate-successor
46
+ preparation/deferral and global cohort rules; local readiness alone is insufficient.
47
+ Count every leaf result. Explicit activations reset activation-local counters,
48
+ not durable slice/recovery state. Pass resolved `--parallel N` to
49
+ [Build](../../rite-build/reference/output.md#caller-owned-continuation).
50
+ Ignored legacy budgets and local serial constraints are not whole-run stops.
51
+ Invented ceilings on dispatches, waits, tool calls or turns cannot stop execution;
52
+ human instructions to pause or limit scope still bind.
55
53
  Resolve ranked validating/blocking picks via `devrites-engine state resolve`;
56
54
  never hand `$rite-resolve` to the user. Escalating, irreversible-risk, access, blocking
57
55
  without recommendations, and Seal judgment (`slice: seal`) still pause.
@@ -74,12 +72,12 @@ Workspace files, not chat, carry state. Read/execute each phase skill:
74
72
  | 10 | `$rite-seal` | GO/NO-GO; no Git |
75
73
  | 11 | `$rite-ship` | only after GO; `--ship` never authorizes Git; stop at literal-GO/native approval |
76
74
 
77
- At every phase entry, before cold/local recovery, on reviewer/planning returns and
78
- before native waits, Autocomplete executes the bounded
79
- [global frontier pass](../../devrites-lib/reference/parallel-workspaces.md#continuous-global-frontier).
80
- Keep source-local return cursors and pending handles; apply its protected
81
- validation window before accepting accounts or landing work. This scheduling
82
- precedes owner Seal and does not change ordered handoff below.
75
+ At phase entry, before local recovery or waits, execute the
76
+ [global pass](../../devrites-lib/reference/parallel-workspaces.md#continuous-global-frontier).
77
+ Consume the fresh frontier emitted after a recorded `dispatch return`; after
78
+ landing, planning/artifact changes or capacity updates, rerun frontier. Execute
79
+ authorized preparation/selection, preserving pending handles, local cursors and
80
+ protected validation windows. Reporting or creating a placeholder is not progress.
83
81
 
84
82
  Check [stop-conditions.md](stop-conditions.md) before advancing. Discard stale
85
83
  pass evidence after source edits. Re-read the selected source before each phase per [phase target binding](../../devrites-lib/reference/parallel-workspaces.md#phase-target-binding). On a cold resume or
@@ -92,19 +90,14 @@ floor; `--update` never excuses an unreconciled regression.
92
90
 
93
91
  ## Sequence continuation
94
92
 
95
- Unless `continue_sequence: false`, while sequence budget remains
96
- ([sentinel](../../devrites-lib/reference/standards/afk-hitl.md#sequence-continuation-continue_sequence-max_workspaces)),
97
- step 10 does not end the run: leave the sealed workspace unshipped with its
98
- `Next step: $rite-ship`, then invoke the next recorded continuation
99
- (`$rite-spec <parent>-<n> "<objective>"` from the parent's `decisions.md` sequence)
100
- and re-enter step 1. Spend one workspace per opened continuation.
101
-
102
- For an already prepared child, do not re-specify. Follow
103
- [`parallel-workspaces.md`](../../devrites-lib/reference/parallel-workspaces.md)
104
- while the owner is still running: honor claims, repair/re-Vet stale readiness,
105
- and admit only explicitly independent/dependency-satisfied isolated work.
106
- ACTIVE changes to the child only at normal ordered handoff after the owner Seal;
107
- resume its durable cursor without a second controller or another opening charge.
93
+ With a recorded successor, opening budget and no opt-out, Seal GO stays internal:
94
+ leave the owner sealed/unshipped with `$rite-ship`, then resume the prepared child
95
+ or invoke its authorized native Spec. Sequence authority follows the
96
+ [shared grammar](../../devrites-lib/reference/workspace-artifact-schema.md#what-each-file-owns),
97
+ not invented slugs. Charge an opening once; preparation may pipeline before Seal,
98
+ one immediate successor per admitted source, not the whole roadmap at once.
99
+ Reuse the child's durable cursor and controller. ACTIVE changes only at normal
100
+ ordered handoff; early planning or isolated independent Build does not advance it.
108
101
 
109
102
  Before invoking a continuation, require durable `state.md`
110
103
  `sequence_workspaces_remaining > 0`; if absent, first write
@@ -158,18 +151,19 @@ and park it after three escalations while independent work continues.
158
151
 
159
152
  ## Continuous caller obligation
160
153
 
161
- Green slice/phase checkpoints consume their recorded `next_action` internally;
162
- only the caller replies finally at completion or a genuine stop.
163
- Before a reply, `devrites-engine check continuation` must print `continuation: clear`; `required` means invoke `invoke:` now.
164
- No user-facing reply is permitted while durable state contains agent-owned
165
- `NEEDS_REPLAN`, an intermediate Plan/Vet action, a distinct retained
166
- Critical/Important fingerprint below its cap or with an escalation left, runnable
167
- work independent of a parked fingerprint, or an open blocking question that
168
- already names a ranked recommended option. Invoke the next internal repair or
169
- `devrites-engine state resolve` immediately, unless that writer already hit the
170
- [resolver-failure stop](stop-conditions.md#other-stop-classes). **Failing case:**
171
- `state resolve` exits 1 on every call and the loop re-invokes it without end. A narrow reviewer closing one finding and exposing another
172
- Critical/Important invariant is progress, not exhaustion.
154
+ After a green slice or phase, execute its recorded `next_action` internally;
155
+ reply only at completion or a genuine stop.
156
+
157
+ Before replying, run `devrites-engine check continuation`: `required` means
158
+ consume `invoke:` now; only `clear` permits a final reply. Under controller mode,
159
+ actionable normalization, successor preparation or foreign work precedes local
160
+ Build. Local-only reconciliation, repair, picks or discharged scheduling deferrals
161
+ retain the owning phase edge, not recursive `$rite-autocomplete`.
162
+ Agent-owned NEEDS_REPLAN, budgeted Critical/Important recovery, independent work
163
+ or a recommended blocking answer stays internal. Invoke repair or `state resolve`
164
+ unless the [resolver-failure stop](stop-conditions.md#other-stop-classes) applies.
165
+ A distinct invariant is progress, not exhaustion; repeated failed resolver calls
166
+ without correction are not progress.
173
167
 
174
168
  The number of completed repair/Vet cycles, context pressure, compaction, session
175
169
  duration, and nested completion never turn checkpoints into handoffs. Persist and
@@ -28,24 +28,27 @@ Fully escalated fingerprints park unless remaining owner is human.
28
28
 
29
29
  ## Not stops: internal technical routing
30
30
 
31
- AFK + `status=running` + an internal phase `next_action` is not a stop. `devrites-engine check continuation` prints `required` for that cursor and `clear` for `$rite-ship`, `awaiting_human`, a question cursor, or an open blocking or escalating question.
32
-
33
- `NEEDS_REPLAN` is not a stop condition under active Autocomplete. It blocks
34
- forward work, persists the return cursor, and invokes Plan repair plus Recovery
35
- Vet inline. The same applies to nested `STOP`/`Next step` when its decision is
36
- agent-owned.
37
-
38
- Agent-owned backtracking is not a stop condition while its exact causal
39
- fingerprint has budget: invoke earlier phase, affected Vet, remediation, proof;
40
- resume origin. Three no-progress corrections spend only that fingerprint's current
41
- strategy: escalate inline to the next class (shared root cause, diagnostic
42
- amplification, boundary redesign) with fresh budget, without permission. After
43
- three escalations per activation it parks; finish independent work, then stop once.
44
- Preserve reproduction, attempts, and dead ends; never offer `$rite-plan unblock`
45
- or another routine phase command. A new user-issued `$rite-autocomplete`
46
- reopens parked fingerprints with fresh escalation; wakes and compaction do not.
47
- Closure is progress; a separately evidenced Critical/Important invariant starts
48
- its own cap and cannot extend the closed one.
31
+ For AFK `status=running`, consume an internal phase edge. In
32
+ `mode: autocomplete`, actionable global normalization, preparation or foreign
33
+ work takes priority over local Build; local-only picks do not recurse through
34
+ Autocomplete. Source human/Ship holds remain binding, while independent
35
+ authorized planning can continue. `clear` is reply permission, not Ship approval.
36
+
37
+ A preparation refusal routes native work or a narrow disclosed scheduling
38
+ deferral under the [global pass](../../devrites-lib/reference/parallel-workspaces.md#continuous-global-frontier).
39
+ Deferral removes only its named scheduling obligation from repeated continuation:
40
+ it neither makes a placeholder prepared nor waives readiness, malformed authority,
41
+ claims, isolation, proof or human gates. Qualified foreign serial handoff is a
42
+ separate rule. Revisit deferrals when their constraint changes.
43
+
44
+ Agent-owned NEEDS_REPLAN or nested STOP persists the return cursor and invokes
45
+ Plan repair, affected Recovery Vet, remediation and proof inline. Three
46
+ no-progress corrections spend that fingerprint's strategy; escalate through
47
+ shared root cause, diagnostic amplification and boundary redesign without asking.
48
+ After three escalations, park it and finish independent work. Preserve attempts;
49
+ resume through `$rite-autocomplete`, not a routine phase command. A new explicit
50
+ invocation reopens parked fingerprints; wakes/compaction do not. Closure is
51
+ progress; a distinct Critical/Important invariant gets its own cap.
49
52
 
50
53
  Spent consumptive-action authorization is not technical-recovery exhaustion.
51
54
  It blocks another execution; retained new Critical/Important evidence still
@@ -74,14 +77,10 @@ Without a safe seam, escalate or park; only a real human/risk/scope gate stops f
74
77
  - **Clarify:** material `open`/unowned decision coverage or a
75
78
  low-confidence high-consequence assumption. Continue the initial interview;
76
79
  never arm AFK early.
77
- - **Seal:** report remaining NO-GO with every blocker/fix direction; never round
78
- up to GO. Resolve recommended-option blocking questions internally; recover
79
- agent-owned findings and re-enter Seal. A first NO-GO is not a stop:
80
- recover every agent-owned blocker (missing or unbound proof, a reviewer gap root
81
- can close, a finding with a fix direction, a missing `review.md` verdict), then
82
- re-enter Seal. The unattended `N` to accepting Important risks means repair, not
83
- stop. Stop only for blockers that need a human: accepting a risk, changing scope
84
- or an acceptance criterion, or access. Record each one as a question.
80
+ - **Seal:** NO-GO never rounds up to GO. Recover agent-owned missing/unbound proof,
81
+ reviewer gaps and findings, then re-enter Seal; declining Important-risk
82
+ acceptance means repair. A human-owned risk, scope/acceptance change or access
83
+ blocker remains a question. A first repairable NO-GO is not a handoff.
85
84
  - **Reslice:** execute its marked action before deciding continue/stop.
86
85
  - **Slice budget:** do not stop on `--max-slices`. Leftover
87
86
  pre-existing remaining value, explicit flag, sentinel cap, or post-vet pending count
@@ -91,8 +90,9 @@ Without a safe seam, escalate or park; only a real human/risk/scope gate stops f
91
90
  another session blocks only work on those paths; run independent work, re-check,
92
91
  and stop only when nothing independent remains. `released`/`dead` claims are
93
92
  reconciled per [§ Cancellation](../../devrites-lib/reference/parallel-dispatch.md#cancellation-and-terminal-reconciliation).
94
- - **Resource envelope:** overlapping run or missing AFK after arm still stops.
95
- Do not stop on `--max-slices`, `max_agents`, `max_minutes`, or `max_review_queue`.
93
+ - **Resource envelope:** overlapping controllers, missing AFK after arm or invalid/
94
+ duplicate mode fail closed. Ignored slice/agent/time/review ceilings do not stop
95
+ Autocomplete; mode itself grants no authority.
96
96
  - **Confidence:** intent still cannot become testable acceptance after interview.
97
97
  - **Repeated failure:** not a stop. Escalate; fully escalated fingerprints park.
98
98
  Stop only when no independent work remains.
@@ -4,11 +4,11 @@
4
4
 
5
5
  Opt-in only. Default `$rite-build` stays [`one-slice-cycle.md`](one-slice-cycle.md).
6
6
 
7
- Autocomplete first executes shared
8
- [`parallel-workspaces.md`](../../devrites-lib/reference/parallel-workspaces.md):
9
- it owns global admission, authorized source discovery and controller routing.
10
- The local selection below applies to standalone Build; mixed selection uses
11
- that reference and then this same per-slice review/repair/integration lifecycle.
7
+ Autocomplete executes the shared [global pass](../../devrites-lib/reference/parallel-workspaces.md#continuous-global-frontier)
8
+ before local writers: actual immediate-successor preparation or narrow disclosed
9
+ deferral, then global cohort admission. Consume fresh dispatch-return scheduling.
10
+ Standalone Build stays local; select notices never broaden its picks. A placeholder
11
+ is not preparation; one successor pipelines safely without preparing the whole roadmap.
12
12
 
13
13
  ## Entry
14
14
 
@@ -17,13 +17,15 @@ that reference and then this same per-slice review/repair/integration lifecycle.
17
17
  | omitted / `--parallel 1` | Serial one-slice |
18
18
  | `--parallel N` (**2≤N≤10**) | N is a **cap**, not a quota: run the largest eligible set ≤ N |
19
19
  | non-integer / `N≤0` / `N>10` | Hard refuse (no silent clamp) |
20
- | Autocomplete `--parallel N` | N is the cap for this run; leftover sentinel `max_parallel` is ignored, then `$rite-autocomplete` writes or replaces only `max_parallel: N` |
20
+ | Autocomplete `--parallel N` | Global cap for this run; controller entry upserts mode and explicit capacity, preserving other AFK fields |
21
21
  | Autocomplete, no flag | Same path, cap from sentinel `max_parallel` else 10; serial only when ineligible |
22
22
 
23
23
  AFK caps `N`. Charge each green sibling once after its successful landing;
24
24
  rejected/unlanded work charges zero, without undoing prior landed accounting.
25
25
  `parallel create` refuses another retained repository controller.
26
- Reconcile retained repair-round artifacts by exact causal fingerprint before honoring a stored blocked status/verdict. Blocked status, round count, or distinct finding alone cannot stop cold resume: establish missing or stale review coverage under step 4; repair unless a stop condition remains evidenced.
26
+ Cold resume reconciles retained repair artifacts by exact fingerprint. A blocked
27
+ label, round count or distinct finding alone does not stop: establish current
28
+ review coverage under step 4 and repair unless a genuine stop remains.
27
29
 
28
30
  ## Dynamic selection and re-batching
29
31
 
@@ -89,14 +91,14 @@ Before `parallel select`, add each slice's hidden write surfaces (list owner:
89
91
  snapshot directory, formatter sweep — and compare paths case-folded
90
92
  (`src/Api.ts` = `src/api.ts`); `--from-plan` adds the manifest/lockfile for a package.
91
93
  Slices colliding on either run serially; the rest
92
- stay eligible. Record `serial: <ids> — <surface>` in `evidence.md`. Under AFK
93
- with `max_parallel ≥ 2`, `dispatch open` refuses a build wright wave narrower
94
- than the from-plan batch until that wave holds the other picks or `serial:`
95
- names each skipped one. Repairing a slice, or a slice already in the lease,
96
- does not waive them. A `serial:` reason that only says the skipped slice is
97
- repairing, already started, or in a wave does not count. HITL is never refused. **Failing
98
- case:** two slices each add a dependency, run in parallel, and integrate fails or
99
- lands a broken lockfile.
94
+ stay eligible. Record `serial: <ids> — <surface>` in `evidence.md`. AFK cap≥2
95
+ refuses narrower Build-wright waves until other picks are started/reserved or
96
+ each skipped pick has a substantive serial reason. Repairing/already-started/in-wave
97
+ alone is not a reason. Controller admission also includes ready foreign picks:
98
+ only exact `serial: child/SLICE-001 — <reason>` can defer that foreign lane; bare
99
+ `SLICE-001` is local. Preparation `frontier-blocked:` cannot waive ready Build
100
+ picks or safety gates. Qualified unsupported-isolation deferral preserves truthful
101
+ owner dispatch/ordered handoff, never early same-tree foreign Build. HITL stays unchanged.
100
102
 
101
103
  ## Control vs workers
102
104
 
@@ -161,8 +163,8 @@ means controller; CLI lease `--slice` uses source-qualified foreign keys.
161
163
  (imperative summary, never a slice id). Staging branch, durable exact landing
162
164
  intent, then FF; retry the same operation after interrupted persistence.
163
165
  The engine reconciles only that recorded landing and refuses unrelated
164
- HEAD/ref movement. Conflict → `integrate-failed`; classify under repair rounds
165
- (siblings stay in place).
166
+ HEAD/ref movement. Conflict → `integrate-failed`; repair in place, retaining
167
+ slot/paths while independent safe landings/refill continue.
166
168
  Never integrate because a wright returned.
167
169
  7. Each landing: FF control (the branch that held `HEAD` when the batch started —
168
170
  often `main`/`master`, otherwise whatever the user was on) by one local commit,
@@ -12,66 +12,45 @@ candidate bindings and proof belong to that source. A missing source artifact
12
12
  routes its prerequisite; it never falls back to another workspace's artifacts.
13
13
  An explicit valid workspace does not require ACTIVE to name it or to exist.
14
14
 
15
- Carry that source through Spec → Clarify → Temper → Define/Plan → Vet and later
16
- Build/Converge/Prove/Polish/Review/Seal/Ship entrypoints. Use existing slug arguments
17
- where supported, including `context <source>` and `handoff <source> --resume`.
18
- For qid-only `state resolve` (including `--drop`/`--batch`), keep its existing argv
19
- and bind the call with `DEVRITES_ROOT=<absolute-control>/.devrites` and
20
- `DEVRITES_WORKSPACE=<absolute-control>/.devrites/work/<source>`. The override is a
21
- workspace path, not a slug. Apply this environment only to that invocation;
22
- neither a fabricated resolver slug argument nor process-global environment
23
- mutation is routing. Use the same source binding for nested recommended answers.
24
-
25
- Preparing or running an admitted future source preserves the ordered owner's
26
- ACTIVE and retained writer/controller. Ordinary standalone Spec still sets ACTIVE;
27
- explicit `--lookahead` still stops at planning/Vet and never becomes a Build root.
15
+ Carry SOURCE through every phase, nested resolution and cold handoff. Use existing
16
+ slug arguments (`context <source>`, `handoff <source> --resume`). Qid-only
17
+ `state resolve` retains its argv: bind only that invocation's `DEVRITES_ROOT` and
18
+ `DEVRITES_WORKSPACE` as above, never a fabricated resolver slug or process-global
19
+ change. Early future work preserves ACTIVE and the retained controller.
20
+ Standalone Spec still sets ACTIVE; `--lookahead` remains planning/Vet-only.
28
21
  Target binding changes neither phase approval nor Ship authority.
29
22
 
30
23
  ## Authority and admission
31
24
 
32
- Resolve one controlling root and ordered lifecycle owner. `.devrites/ACTIVE` stays
33
- on that owner until normal sequence handoff; scheduling a future slice does not
34
- advance ACTIVE or create a second Build root. One repository-global lease owns
35
- control-tip mutations. Resume its controller, source identities and retained
36
- handles before admitting more work; status queried via ACTIVE/source reports the
37
- actual retained controller even after ordered handoff. Another retained
38
- controller is not bypassable; no extra cursor pointer is needed.
39
-
40
- The existing `--parallel N` is one global active-writer cap across all sources:
41
- explicit invocation value, else AFK `max_parallel`, else retained lease capacity
42
- on resume, else 10. Explicit 1 means one writer, not one writer per workspace.
43
- Review queues obey host capacity and existing reviewer caps separately; they never
44
- erase required specialist accounts.
45
-
46
- Discover only recorded forward sequence continuations through `sequence_parent`
47
- and the parent's recorded objectives. Honor `continue_sequence: false`, opening
48
- budget and human bounded requests. Charge each opening once; reuse prepared
49
- children and their cursor rather than re-specifying or charging them again.
50
- An unrelated directory or user-mentioned slug is not automatic enrollment.
51
- Respect workspace claims: a live lookahead owner retains its artifacts until
52
- release/terminal reconciliation. Root alone writes workspace artifacts; leaves
53
- never write `.devrites/**`.
54
-
55
- For each future slice, native Plan/Vet must have CURRENT approved readiness,
56
- a complete bound contract and explicit `Workspace dependencies` under
57
- [canonical slice grammar](workspace-artifact-schema.md#canonical-slice-grammar).
58
- Plan/Vet checks assumptions, consumed interfaces, shared runtime resources and
59
- independent slice proof. `none` is a native declaration, never an inference from
60
- path disjointness, local `Dependencies: none`, planning digests or a green branch.
61
- Only durably BUILT/integrated producer checkpoints satisfy qualified dependencies;
62
- validate each producer reference and graph, including safely archived predecessors.
63
- Cycles, unknown/malformed references and shared unisolated runtime withhold unsafe
64
- candidates. A waiting producer withholds only its consumers, not unrelated work.
65
-
66
- Missing declarations in legacy future plans need narrow Plan repair + affected
67
- Vet before early admission; existing local/standalone plans remain valid. Absent,
68
- incomplete, stale or human-blocked future work is a planning/repair/blocked
69
- opportunity, not a fatal pool failure. Retain every readiness reason: fixing stale
70
- bindings does not waive a coexisting human/access/safety gate. Re-Vet affected
71
- predecessor assumptions, then record current bindings; never auto-refresh a digest.
72
- Read archived predecessors in place through the engine's referenced read-only
73
- readiness lookup. Never copy an archive into live `work/` merely to read/hash it:
74
- that can resurrect stale controller leases and claims.
25
+ Keep ACTIVE on the ordered owner until normal handoff. One repository-global
26
+ lease owns control-tip mutations; recover its controller, source identities and
27
+ handles before more admission, including after handoff. No second Build root.
28
+ `--parallel N` caps active writers globally: explicit N, else AFK `max_parallel`,
29
+ else retained capacity on resume, else 10. One means one writer total. Review
30
+ queues retain host/reviewer caps; capacity or Reslice never waives fanout or proof.
31
+
32
+ Only checked source-related sequence membership qualifies; user-mentioned or
33
+ unrelated directories are not enrollment. Current canonical `roadmap.md` follows
34
+ [the grammar](workspace-artifact-schema.md#what-each-file-owns): one related live/
35
+ archived owner, actual chain root, same-owner unique decision and contiguous
36
+ position table. It supersedes historical tables for scheduling without rewriting
37
+ them. Invalid canonical authority fails closed; absence retains conservative
38
+ legacy resolution. Honor opt-out, opening budgets and human bounds; charge each
39
+ opening once. Lookahead claims retain artifacts through terminal reconciliation.
40
+ Root alone writes workspace artifacts; leaves never write `.devrites/**`.
41
+
42
+ Native Plan/Vet must bind CURRENT readiness, contract, producer assumptions and
43
+ explicit future `Workspace dependencies` per [slice grammar](workspace-artifact-schema.md#canonical-slice-grammar).
44
+ `none` is a native declaration, not an inference from paths, local dependencies,
45
+ digests or green branches. Only durable BUILT/integrated checkpoints satisfy
46
+ validated producer references/graphs, including safely archived predecessors.
47
+ Unknown references, cycles and shared unisolated runtime withhold unsafe slices;
48
+ waiting producers withhold consumers, not independent work. Legacy future plans
49
+ need narrow Plan repair/affected Vet; standalone/local plans retain their rules.
50
+ Per-slice withheld diagnostics do not block other returned eligible IDs;
51
+ workspace-wide readiness failures still withhold the source.
52
+ Preserve coexisting stale-readiness and human/access/safety reasons. Re-Vet changed
53
+ assumptions; never auto-refresh digests or copy archives into live `work/`.
75
54
 
76
55
  ## Recover avoidable serialization
77
56
 
@@ -109,64 +88,77 @@ reslicing unchanged work.
109
88
 
110
89
  ## Continuous global frontier
111
90
 
112
- Autocomplete executes this bounded pass at every phase entry (including cold
113
- Prove/Vet/Review), before phase-local recovery, on each landing/planning/review
114
- return, and immediately before a native wait or final reply. A nested phase
115
- returns scheduling to Autocomplete; its source-local phase/recovery return cursor
116
- and pending native reviewer handles remain intact. This is not a second phase
117
- controller or a persistent frontier artifact.
118
-
119
- 1. Re-read controller lease, per-source state/questions/readiness and authorized
120
- sequence. Examine every recorded authorized source once per pass, not only
121
- the owner's next local slice or first blocked child. Prepare absent targets
122
- through authorized native Spec → Clarify → Temper → Define/Plan → Vet; repair
123
- and re-Vet current readiness of prepared targets now, not after owner Seal.
124
- Assess actual source gates, claims and explicit planning/Build restrictions
125
- before routing. Owner-local red/human/access gates withhold that owner or its
126
- actual consumers, not independent authorized planning; true whole-run safety,
127
- access and claim constraints still bind. Retain concrete actions or blockers
128
- in existing source artifacts under normal write authority.
129
- Yield a child at its pending reviewer/phase boundary back to this controller;
130
- never recursively wait for child completion while parent returns are pending.
131
- Reconcile available parent returns before another pass. A native wait is
132
- justified only after concrete authorized actions/blockers across the pool
133
- have been examined; reading later phase skills is not pending Plan/Vet work.
134
- If an explicit/AFK capacity change differs from retained capacity, reconcile it
135
- before selection using `parallel add --root <repo> --slug <controller> --cap N --json -`
136
- with stdin `[]`. This changes capacity only, not workers or a complete lease's
137
- status. A decrease below active count refuses: pause admissions and drain to
138
- the requested count, then retry; never cancel workers to meet a lower cap.
139
- A pending landing is recovery, not free capacity: retry the exact recorded
140
- integrate before admission; unrelated HEAD/ref movement remains blocked.
141
- 2. Query the mixed pool in stable sequence order, controller first (the ordered
142
- ACTIVE owner may now differ):
143
- `devrites-engine parallel select --root <repo> --slug <controller> --from-plan --workspaces <controller,authorized-source,...> --cap N`.
144
- `--workspaces` is an internal data-source parameter, not a user feature flag.
145
- Read all notices and route repair/waits to their source. Existing running slots
146
- and paths seed selection globally. Zero new picks with live writers means keep
147
- planning/reviewing or wait, not Prove for those unfinished sources.
148
- 3. Preflight selected slices against their SOURCE workspace with Build's normal
149
- slice/readiness/precondition, exact-path and runtime-isolation gates. Early
150
- foreign Build always uses isolated workers. For cap>1 on a capable host,
151
- isolate even the first lone local writer, so later sources can refill. Persist
152
- `--cap N` on create/add. With a retained lease, add even one pick; a complete
153
- retained lease can reopen only at unchanged control tip. After cleanup,
154
- create a new batch from current control. Never start another controller.
155
- Unsupported isolation uses truthful serial owner dispatch and ordinary ordered
156
- handoff, never early same-tree foreign Build or concurrent same-worktree writers.
157
- 4. Launch every selected native writer before waiting on any; keep scheduling
158
- planning and independent review while they run, including while the owner
159
- awaits Prove/Vet/Review accounts. Apply the protected window below before any
160
- landing: defer control mutations while frozen validation is pending. Outside
161
- that window, land each individually green sibling through Build review/proof
162
- and `parallel integrate --slice <key>`; refill immediately from the whole pool.
163
- A red sibling repairs in place and holds its slot/paths; it does not delay
164
- independent safe landings. Never satisfy a producer edge on writer return or
165
- green-but-unlanded work.
166
- After worker-bound review/proof and source bookkeeping finish, prune only
167
- verified clean integrated checkouts with `parallel prune --root <repo>
168
- --slug <controller>`. Retain transfer refs/lease identities; dirty or ignored
169
- data stays with its reported recovery path and does not justify forced deletion.
91
+ Run this pass at phase entry, before local recovery/waits/reply, and after landing
92
+ or planning changes. A persisted `dispatch return` emits fresh scheduling after
93
+ locks release, even when later claim release fails: retain the nonzero failure
94
+ and claim exclusion. Exact successful claim reconciliation also refreshes;
95
+ rejected/unpersisted returns do not. Ordinary status stays read-only. Keep local
96
+ return cursors and native handles: one controller, not another phase or queue.
97
+
98
+ 1. Read `devrites-engine frontier <source>`, then its selected `frontier-authority`
99
+ paths before preparation. Follow the actual owner/decision, including archives;
100
+ local phase context is not a reason to create a second owner. Execute native actions.
101
+ Normalize unresolved legacy sequence authority through its owning decision/
102
+ phase. Positive budget, ambiguous prose or path disjointness grants nothing.
103
+ A current invalid canonical map is an error, never a fallback to history.
104
+ For controller mode at cap>1, before a new Build-wright wave writes claims,
105
+ rounds or history, start preparation of that source's immediate recorded
106
+ successor. Missing targets need native Spec → Clarify → Temper → Define/Plan →
107
+ affected Vet; existing targets need completed native state, actual pending
108
+ planning/review handles, current Build-ready binding, or a human/dependency hold.
109
+ Completed live/archived successors need no reopening; retained actors still bind.
110
+ A placeholder, prose claim, unopened wave or unrelated reviewer is insufficient.
111
+ Pipeline one successor per admitted source; preparing the whole roadmap first
112
+ is unnecessary. Opt-out, zero opening budget or source human/question holds
113
+ prevent new openings, not already-created child planning or independent work.
114
+ If preparation cannot start, retain `frontier-blocked: <successor> — <constraint>`
115
+ in source `evidence.md`. `frontier-blocked: sequence — <constraint>` may defer
116
+ unresolved historical normalization for valid independent local work. Each row
117
+ names one concrete native-reviewed constraint; keep it visible and revisit on
118
+ change. It discharges only that scheduling obligation, not readiness, malformed
119
+ canonical/AFK data, paths, claims, isolation, proof or human gates. It cannot
120
+ waive already-ready foreign picks or certify preparation/normalization.
121
+ 2. Reconcile explicit/AFK capacity via `parallel add --root <repo> --slug
122
+ <controller> --cap N --json -`, stdin `[]`. Below occupied count, freeze every
123
+ new launch, even reserved waves; reconcile/drain and retry without cancelling.
124
+ Recover recorded landing intent; unrelated HEAD/ref movement
125
+ stays blocked. Rerun frontier after changes and execute its exact authorized
126
+ `frontier-select-command`. Local select notices do not broaden picks. Existing
127
+ global slots and paths seed selection; reserve the mixed cohort in create/add
128
+ before separate source-local wave opens. Reservations assemble width, never
129
+ prove actor checkout/start. Ready foreign picks cannot disappear behind local
130
+ `serial:` records. A foreign deferral is exactly
131
+ `serial: child/SLICE-001 — <reason>` in source evidence; bare `SLICE-001` is local.
132
+ Substantive-reason rules apply; native workflow judges the constraint.
133
+ Re-evaluate owning serial rows when constraints change; engine checks exact
134
+ scope, not the truth of a host/resource claim.
135
+ Discharged deferrals are not repeatedly actionable. Foreign work/preparation
136
+ and authority normalization precede local Build; local-only reconciliation,
137
+ repair and picks retain their owning phase, not recursive Autocomplete.
138
+ Zero picks with live writers means planning/wait, not premature Prove.
139
+ 3. Apply SOURCE Build readiness/preconditions, exact paths and runtime isolation.
140
+ Early foreign Build always uses isolated workers. On a capable host at cap>1,
141
+ isolate even the first local writer for later refill; persist `--cap N`.
142
+ Add one pick to a retained lease; reopen a complete lease only at unchanged
143
+ control tip, or create from current control after cleanup. Unsupported
144
+ isolation uses qualified serial deferral, truthful local-owner dispatch and
145
+ ordinary ordered handoff, never early same-tree foreign/concurrent writers.
146
+ New dispatch accounts retain opaque `controller_batch` and immutable per-slice
147
+ normalized `admitted_paths`. Retain only each pending writer's lane; returned/
148
+ integrated siblings may release slots for rolling refill. Missing snapshots or
149
+ changed lease paths require actor reconciliation, never backfill from new tasks.
150
+ Exact claim identity/session still matters; release/expiry does not stop an actor.
151
+ Same-session claims allocate space, not extra write authority. None of these
152
+ records proves actual cwd/profile: verify through the native host.
153
+ 4. Start every selected native writer before waiting; continue independent native
154
+ planning/review while they run. Protected validation defers unsafe landings,
155
+ not independent preparation. Outside that window, land individually green
156
+ siblings through Build review/proof and `parallel integrate --slice <key>`;
157
+ refill globally. Red/integrate-failed siblings keep slots/paths for repair.
158
+ Durable integration alone satisfies producer edges. After worker-bound
159
+ accounts and source bookkeeping settle, `parallel prune --root <repo> --slug
160
+ <controller>` removes only verified clean integrated checkouts, retaining
161
+ transfer refs/lease identities. Dirty/ignored data keeps its recovery path.
170
162
 
171
163
  ## Identity, dispatch and proof
172
164