open-multi-agent-kit 0.98.5 → 1.0.0

package/dist/modes/interactive/resource-description.d.ts

@@ -0,0 +1,3 @@
+/** Keep decorative imported harness markers out of OMK display; source metadata stays unchanged. */
+export declare function formatResourceDescription(description: string | undefined, sourceTag?: string): string | undefined;
+//# sourceMappingURL=resource-description.d.ts.map

package/dist/modes/interactive/resource-description.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"resource-description.d.ts","sourceRoot":"","sources":["../../../src/modes/interactive/resource-description.ts"],"names":[],"mappings":"AAAA,oGAAoG;AACpG,wBAAgB,yBAAyB,CAAC,WAAW,EAAE,MAAM,GAAG,SAAS,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAWjH","sourcesContent":["/** Keep decorative imported harness markers out of OMK display; source metadata stays unchanged. */\nexport function formatResourceDescription(description: string | undefined, sourceTag?: string): string | undefined {\n\tlet text = description;\n\tif (description) {\n\t\tconst marker = /^\\[(OMX|OMO)\\]\\s*/.exec(description);\n\t\tif (marker) {\n\t\t\tconst body = description.slice(marker[0].length);\n\t\t\ttext = body || \"OMK resource\";\n\t\t}\n\t}\n\tif (!sourceTag) return text;\n\treturn text ? `[${sourceTag}] ${text}` : `[${sourceTag}]`;\n}\n"]}

package/dist/modes/interactive/resource-description.js

@@ -0,0 +1,15 @@
+/** Keep decorative imported harness markers out of OMK display; source metadata stays unchanged. */
+export function formatResourceDescription(description, sourceTag) {
+    let text = description;
+    if (description) {
+        const marker = /^\[(OMX|OMO)\]\s*/.exec(description);
+        if (marker) {
+            const body = description.slice(marker[0].length);
+            text = body || "OMK resource";
+        }
+    }
+    if (!sourceTag)
+        return text;
+    return text ? `[${sourceTag}] ${text}` : `[${sourceTag}]`;
+}
+//# sourceMappingURL=resource-description.js.map

package/dist/modes/interactive/resource-description.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"resource-description.js","sourceRoot":"","sources":["../../../src/modes/interactive/resource-description.ts"],"names":[],"mappings":"AAAA,oGAAoG;AACpG,MAAM,UAAU,yBAAyB,CAAC,WAA+B,EAAE,SAAkB,EAAsB;IAClH,IAAI,IAAI,GAAG,WAAW,CAAC;IACvB,IAAI,WAAW,EAAE,CAAC;QACjB,MAAM,MAAM,GAAG,mBAAmB,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QACrD,IAAI,MAAM,EAAE,CAAC;YACZ,MAAM,IAAI,GAAG,WAAW,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;YACjD,IAAI,GAAG,IAAI,IAAI,cAAc,CAAC;QAC/B,CAAC;IACF,CAAC;IACD,IAAI,CAAC,SAAS;QAAE,OAAO,IAAI,CAAC;IAC5B,OAAO,IAAI,CAAC,CAAC,CAAC,IAAI,SAAS,KAAK,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,SAAS,GAAG,CAAC;AAAA,CAC1D","sourcesContent":["/** Keep decorative imported harness markers out of OMK display; source metadata stays unchanged. */\nexport function formatResourceDescription(description: string | undefined, sourceTag?: string): string | undefined {\n\tlet text = description;\n\tif (description) {\n\t\tconst marker = /^\\[(OMX|OMO)\\]\\s*/.exec(description);\n\t\tif (marker) {\n\t\t\tconst body = description.slice(marker[0].length);\n\t\t\ttext = body || \"OMK resource\";\n\t\t}\n\t}\n\tif (!sourceTag) return text;\n\treturn text ? `[${sourceTag}] ${text}` : `[${sourceTag}]`;\n}\n"]}

package/docs/audit-revalidation-2026-09-17.md

@@ -0,0 +1,77 @@
+# Audit revalidation — 2026-09-17
+
+Baseline: `1e58611d0c28775c8a066b6e913ea6710ccea396`, with unrelated local
+provider/model changes present. This is a scoped local revalidation, not a release
+approval, full security audit, or competitor benchmark.
+
+## Inputs and scope
+
+The local September 14 reproduction ZIP targets
+`69e0637d8fddd26eb40a7ec3f8ebfb65b500de5b`. All 25 entries listed in its
+SHA256SUMS matched. Its recorded six expected failures describe that historical
+snapshot, not the current checkout. The archived runner was not executed or
+substituted for current repository tests.
+
+The September 15 strategy report combines historical findings with proposed
+resource-aware scheduling, recovery, and comparative evaluation. These proposals
+are not measured product benefits. The September 16 GitHub audit targets
+`739bc6f3b6fe1c89bfe058aad7ec17ad252fb10e`; its missing DAG contract is no longer
+present in this baseline. The documents were used to select the checks below;
+not every proposed acceptance criterion was executed.
+
+## Executed regression groups
+
+Commands use `node ../../node_modules/vitest/dist/cli.js --run` from the named
+workspace. No external provider requests were needed.
+
+| Workspace | Test files | Result |
+| --- | --- | --- |
+| agent | `test/tool-dag-*.test.ts` | 83 passed, including 10,000 seeded bounded schedules |
+| coding-agent | `test/mcp/{protocol,protocol-required,manager-lifecycle,manager-health,client,tools}.test.ts` | 61 passed; client tests include a local stdio fixture |
+| protocol | `test/{protocol,evaluation-candidate-binding,validation,run-dag-contract,run-dag-properties}.test.ts` | 44 passed |
+| coding-agent | `test/{context-budget-v2-validation-cache,context-budget-cache,context-budget-cache-policy,context-budget-v2-tier-floor,context-budget-v2-eligibility,context-budget-v2-nonfinite-tokens,context-budget-governor-v2}.test.ts` | 46 passed after the fix below |
+
+These are 234 distinct passing test cases, not 234 independent production tasks.
+An initial command included nonexistent `test/mcp/manager.test.ts`; Vitest ran
+other matching files successfully. Manager coverage above comes from the actual
+`manager-lifecycle`, `manager-health`, and `client` files, not that missing path.
+
+## Fixed: input diagnostics leaked across plan-cache reuse
+
+`planPromptContextBudgetV2()` originally decided cache eligibility before
+`validateBudgetItems()`. Duplicate IDs are dropped and invalid token estimates are
+recomputed. Their sanitized items can produce exactly the same plan key as valid
+input, but their input diagnostics are different.
+
+Three regressions failed before the fix:
+
+1. A cached valid plan hid a duplicate-ID diagnostic on a later call.
+2. A plan produced from duplicate IDs carried its diagnostic into a later valid call.
+3. A cached plan hid the diagnostic for a recomputed non-finite token estimate.
+
+Cache eligibility is now decided after input validation. Calls with input or
+budget diagnostics neither read nor write the plan cache. Representation-cache
+validation and normal valid-input plan reuse remain unchanged. The regression
+uses a required item to isolate plan reuse from the existing
+`cache_dependency_unsafe` rejection for plans containing representation-cache hits.
+
+Changed implementation: `src/core/context-budget-v2-planner.ts`.
+Regression: `test/context-budget-v2-validation-cache.test.ts` (3 passed).
+This preserves diagnostics; it does not redesign duplicate-item handling or tier
+allocation policy.
+
+## Remaining boundaries
+
+- Full `npm run check` has been blocked by unrelated local module-size growth in
+  `agent-session.ts` and `model-registry.ts`; those files and the ratchet baseline
+  are not changed by this fix. A passing focused compiler/test run does not close
+  the full repository gate.
+- General observation evaluation still has existential semantics. Passing
+  candidate-binding tests does not establish latest-result or complete-coverage
+  semantics, or independently authenticate a verifier.
+- ECRAF remains an internal planner. Resource normalization, fairness, live
+  admission integration, and equal-budget performance comparisons were not added.
+- This pass does not validate the complete timeout/ownership fault matrix,
+  verified-run crash recovery, all MCP authorization boundaries, remote CI,
+  branch protection, published artifacts, or router calibration.
+- No release, commit, push, paid benchmark, or new runtime default is implied.

package/docs/devin-harness.md

@@ -0,0 +1,150 @@
+# Devin SWE-2 harness
+
+This page is the canonical operator guide for the built-in `devin` provider, centered on the logical `swe-2` model. Use `/login devin` for the Devin CLI subscription PKCE flow or `DEVIN_API_KEY` for an already-owned CLI session token. A user-local `~/.omk/agent/devin.md` may add operator notes, but it is not the portable product contract.
+
+Authentication, transport, and verification limits are owned by [Providers](providers.md#devin-cli); this page covers how OMK drives SWE-2 as a harness.
+
+## Presets
+
+Project presets live in `.omk/presets.json` (or `~/.omk/agent/presets.json`) and are consumed by the preset extension from `packages/coding-agent/examples/extensions/preset.ts`. The shared SWE-2 presets intentionally omit the `tools` key so role/domain lane grants keep control of the active tools.
+
+| Preset | Provider | Model | Thinking | Use |
+| --- | --- | --- | --- | --- |
+| `swe2-verified` | `devin` | `swe-2` | `high` | Default SWE-2 coding baseline: multi-file edits with tests. |
+| `swe2-max` | `devin` | `swe-2` | `max` | Long-horizon, uncertain, or repository-wide work that should use the 1M budget. |
+| `swe2-fast-edit` | `devin` | `swe-2` | `medium` | Small, well-specified edits where medium acts sooner and costs less. |
+
+```json
+{
+  "swe2-verified": { "provider": "devin", "model": "swe-2", "thinkingLevel": "high" },
+  "swe2-max": { "provider": "devin", "model": "swe-2", "thinkingLevel": "max" },
+  "swe2-fast-edit": { "provider": "devin", "model": "swe-2", "thinkingLevel": "medium" }
+}
+```
+
+For a new session without presets:
+
+```bash
+omk --provider devin --model swe-2 --thinking max
+```
+
+## Effort tiers
+
+SWE-2 exposes exactly three server-declared efforts. OMK maps its thinking tiers as follows and rejects anything else before sending credentials; `max` is a reasoning level, not the Devin Max subscription tier.
+
+| OMK tier | `swe-2` |
+| --- | --- |
+| `off`, `minimal`, `low` | unavailable |
+| `medium` | `medium` |
+| `high` | `high` |
+| `xhigh`, `ultra` | unavailable |
+| `max` | `max` |
+
+Guidance from the [SWE-2 announcement](https://cognition.com/blog/swe-2): `medium` makes its first real edit sooner and is the cost-efficient choice for simple and intermediate tasks; `high` and `max` plan more, explore more of the codebase, and verify more on complex tasks. `recommendedDevinEffortForIntent()` encodes the same split (`code` → `medium`, `test` → `high`, `debug`/`repo` → `max`).
+
+## Context budget: 1,000,000 tokens
+
+`devin/swe-2` ships with `contextWindow: 1000000` and `maxTokens: 16384`. These are local budgets that drive OMK's context budgeting and compaction, **not published SWE-2 limits**; Cognition has not published a context window for SWE-2. Chat completion settings follow the captured native Devin CLI 3000.6.2 defaults (`maxNewlines` 400, empty stop list). `topP=0.95` is protobuf field 8; field 6 is `firstTemperature` and is omitted. This removes the old 200-newline cap and synthetic stops, but does not prove that every interrupted reply had that cause. The budget also selects the catalog lane:
+
+1. Before each turn OMK reads `GetCliModelConfigs`. SWE-2 family entries may carry a `1M Context` axis (order `1`) beside the effort axis. A local budget of 1,000,000 or more asks for that 1M-context lane; below it, the standard lane is used and 1M entries are ignored.
+2. A catalog with no 1M-context lane keeps the standard lane for the selected effort.
+3. If the chosen lane declares a context window smaller than the local budget, the request fails with `... declares a N-token context window; lower the models.json contextWindow before retrying`. OMK never shrinks the budget silently, never invents a wire UID, and never downgrades to another effort.
+4. Fast-lane (`Fast Mode`) entries are excluded from effort routing and are reachable only through their own UID models (ids ending in `-fast` or `-priority`). Output is capped against the authenticated catalog's declared maximum.
+
+To lower the budget (for example if your account only serves the standard lane at 262,144 tokens), override the built-in model in `~/.omk/agent/models.json`:
+
+```json
+{
+  "providers": {
+    "devin": {
+      "modelOverrides": {
+        "swe-2": { "contextWindow": 262144 }
+      }
+    }
+  }
+}
+```
+
+Recommended compaction settings for 1M sessions keep the defaults but raise the recent-token window so summaries do not discard the working set:
+
+```json
+{
+  "compaction": { "enabled": true, "reserveTokens": 16384, "keepRecentTokens": 60000, "maxUsageRatio": 0.85 }
+}
+```
+
+Context discipline still applies: the budget is room for the repository, not an invitation to dump it. Prefer targeted reads and searches, keep tool output bounded, and let `precompact-checkpoint` snapshot state before compaction rather than restarting sessions. Quota is per account; a larger window consumes more of it per turn.
+
+With a Devin credential configured, the status rail's USAGE section shows the account's daily and weekly quota meters from `GetUserStatus` (or the plan name and credit balances when the plan reports no quota windows), matching the CLI's `/usage` surface.
+
+## Domain routing
+
+Selecting the `devin` provider auto-applies the `devin-harness` loadout by default; this does not require `OMK_DOMAIN_ROUTING=1`. Set `OMK_DEVIN_HARNESS=0` to disable that provider-specific dispatch. The flag is independent from `OMK_GROK_HARNESS`.
+
+Loadout policies reject extension tools that shadow builtins fail-closed. A host that intentionally replaces the builtin `bash` (for example a Landstrip shell provider) should keep `OMK_DEVIN_HARNESS=0` in its launcher until the replacement is removed or bridged; the `devin.md` overlay and the model budget still apply in that case.
+
+General prompt-based domain routing is separate and opt-in through `OMK_DOMAIN_ROUTING=1`. It selects one of the profiles under [`loadout-domains/`](loadout-domains/README.md) and composes it with the active role loadout. SWE-2 presets only set provider, model, thinking level, and instruction pointers.
+
+## Model selection
+
+The `devin` catalog leads with the logical `swe-2` model; the server's SWE-2 family metadata supplies each effort's wire UID at request time. Every other lane the account catalog advertises is its own logical model whose id is the wire UID — for example `devin/claude-opus-5-high`, `devin/gpt-5-6-sol-xhigh`, `devin/gemini-3-8-flash-medium`, `devin/kimi-k3-max`, `devin/glm-5-3-high`, `devin/grok-4-6-xhigh`, `devin/deepseek-v4-pro-max`, `devin/swe-1-7`, or `devin/inkling-max`. Flat models pin their declared effort lane, so `/think` levels are fixed per model and `No Thinking`/`None` lanes report `reasoning: false`. Availability is account- and plan-dependent: a lane absent from your catalog fails loudly instead of being remapped. Use `/model` or `omk --list-models devin` for the current list. Image input is unsupported; provide text.
+
+## Skill and MCP matrix summary
+
+Use the normal OMK lane grant model: grant the smallest skill and MCP surface that matches the task.
+
+For each non-queued `devin` request started through `AgentSession.prompt()`, OMK calls `selectDevinHarnessSkills()` against the live discovered skill descriptions after ordinary prompt-template expansion. It merges up to three matches with explicit/settings selections and rebuilds that turn's `<active_skills source="devin-harness">` marker. The scorer is the same `selectSkills()` used by the Grok harness: weak 0.35 / strong 0.7 thresholds, deterministic input-order ties, first-name-wins deduplication, explicit-only skills never auto-selected, and `headroom` only for lexical pressure cues or the session's measured context-pressure bucket. A task with no signals yields an empty automatic grant rather than the full allowlist. Queued `steer`, `followUp`, or `prompt(..., { streamingBehavior })` messages retain the active run's system prompt.
+
+| Task class | Skills | MCP |
+| --- | --- | --- |
+| Multi-package or repo-context work | `packages`; add `headroom` only under context pressure | none by default |
+| Repo graph or broad comprehension | `understand-anything`; optionally `packages` | `understand-anything` |
+| TypeScript/Rust/Python/Go edits | `programming`; add `lsp` or `ast-grep` only for symbol/structural work | none by default |
+| New behavior or bug fix with regression test | `tdd-workflow`, `programming` | none by default |
+| Runtime failures or broken behavior | `debugging` | task-specific only |
+| Library API lookup | task skill as needed | `context7` |
+| Current public URL or docs lookup | task skill as needed | `fetch` |
+| UI/TUI verification | task skill as needed | `playwright` only when browser/UI evidence is required |
+
+Relevant evidence hooks for SWE-2 lanes are `pre-shell-guard`, `protect-secrets`, `typecheck-after-edit`, `stop-verify`, `session-context`, and `precompact-checkpoint`. Hook output is incremental evidence; code changes still need the project's required final verification command before claiming type/lint cleanliness.
+
+## Suggested TUI flow
+
+1. Run `/login devin` once; the account picker stores the CLI session token in `auth.json`.
+2. Select `/preset swe2-verified` for normal coding work, `/preset swe2-max` for long-horizon or repository-wide tasks, and `/preset swe2-fast-edit` for small edits.
+3. Use `/think medium`, `/think high`, or `/think max` to change effort mid-session; other levels are rejected.
+4. If a turn fails with an "unavailable or ambiguous" route or a smaller declared context window, treat it as a configuration signal: check `devin models list`, or lower `contextWindow` as shown above. Do not retry with a guessed wire UID.
+5. Keep credentials out of preset JSON, prompts, and logs: the session token, the exchanged user JWT, and `auth.json` contents are secrets under `protect-secrets`.
+
+## Troubleshooting `does not provide an export named`
+
+If Devin fails with `The requested module './devin-connect.js' does not provide an export named 'MAX_FRAME_BYTES'`, the loaded stream module is mixed with an older unary module. That is a client load error, not an orphan tool call or a remote protocol trailer.
+
+- Current OMK keeps the 16 MiB Connect frame cap inside `devin-connect-stream.ts`, so a stale `devin-connect.js` cannot fail that named import.
+- `/new` does not reload already-imported provider modules. Quit and restart OMK after a rebuild.
+- The failure banner should say to restart OMK, not to sanitize a sticky transcript.
+
+## Troubleshooting `invalid_argument`
+
+A Connect `invalid_argument` trailer is a provider/request failure, not evidence of
+an orphan tool call or a safety refusal. Keep any reported trace ID for support;
+OMK includes only a bounded hexadecimal trace ID, never the remote error body.
+
+- The SWE-2 adapter rejects zero, negative, and non-finite `temperature` values
+  before sending credentials. A controlled high-effort probe returned
+  `invalid_argument` at `temperature: 0`, while the default temperature completed
+  the same prompt. Omit the option to use the native default of `1`; OMK does not
+  silently replace an explicit zero.
+- Check model, effort, context budget, and request settings with `/debug`. Repair
+  tool history only when a tool/message mismatch is actually identified.
+- After updating or rebuilding the adapter, quit and restart the OMK process.
+  `/new` replaces conversation state; it does not reload an imported provider
+  module. This is an update-application step, not a guaranteed fix for every
+  provider error.
+- If a minimal tool-free request still fails at the default temperature, preserve
+  the trace ID and investigate provider compatibility or availability rather than
+  repeatedly sanitizing the same transcript.
+
+## Local overlay
+
+When the `devin` provider is active, OMK appends `~/.omk/agent/devin.md` (capped at 24,000 characters) to the system prompt, mirroring the Grok `grok.md` overlay. Treat that file as optional host configuration for effort defaults, compaction notes, or team conventions; this page and the current provider documentation remain authoritative and it cannot override higher-priority instructions.

package/docs/docs.json

@@ -27,6 +27,10 @@
           "title": "Native xAI Grok",
           "path": "grok-harness.md"
         },
+        {
+          "title": "Devin SWE-2",
+          "path": "devin-harness.md"
+        },
         {
           "title": "Containerization",
           "path": "containerization.md"

package/docs/ecraf-normalization.md

@@ -0,0 +1,83 @@
+# ECRAF dimensionless normalization (R11)
+
+Status: **pure planner, opt-in, not wired** — the same gate vocabulary as
+[runtime-algorithms](./runtime-algorithms.md) applies. This page documents what is
+implemented and verified, and states explicitly what remains unproven.
+
+## Versions
+
+`planEcrafAdmissions()` accepts an explicit `algorithmVersion`:
+
+| Version | Density formula | When selected |
+| --- | --- | --- |
+| `legacy-v1` (default when omitted) | `P_i / (epsilon + Σ_r weight_r · a_ir)` | No normalization options supplied. |
+| `normalized-v2` | `P_i / (epsilon + slotCost + Σ_r weight_r · a_ir / s_r)` | `algorithmVersion: "normalized-v2"` **or** `referenceScales` supplied without a version (compatibility with the earlier scales-only opt-in). |
+
+Compatibility policy:
+
+- Omitting `algorithmVersion` and every normalization option reproduces the
+  byte-for-byte legacy plan. Legacy regression and property tests are unchanged.
+- Supplying `referenceScales` (or `slotCost`) with `legacy-v1` is rejected with
+  `RangeError`; the scales-only shape silently choosing a different algorithm was
+  the ambiguity this version policy closes.
+- Unknown versions and unknown future options are rejected, not coerced.
+
+## normalized-v2 semantics (spec §13.2)
+
+- Every resource with a nonzero demand needs a positive finite reference scale
+  `s_r`. In `normalized-v2` omitted entries default to the resource's **positive
+  total capacity** — never remaining headroom — so the caller cannot implicitly
+  re-scale ranking by scheduling pressure. Unbounded resources (capacity omitted)
+  cannot fall back to a scale and require an explicit one.
+- Zero capacity is a **feasibility gate**, not a scale: a positive demand on a
+  zero-capacity resource is deferred before scoring; a zero demand still uses the
+  resource in the legacy sense (admission fails on held usage), so the documented
+  "missing capacity = unbounded" and `capacity: 0` meanings are preserved.
+- `slotCost` (λ_slot) is finite and **positive**; the slot term charges every
+  running candidate one execution slot even when its resource vector is empty or
+  all-zero, so an empty node cannot dominate on `epsilon` alone.
+- Infinite capacities are rejected in v2 (`RangeError`); the legacy finite-input
+  contract is unchanged. Finite inputs can still overflow the denominator,
+  density, or reserved usage, and those derived values are rejected with
+  `RangeError` as before.
+
+## Verified properties (unit invariance, spec §13.3)
+
+`a'_ir = c_r·a_ir` with `s'_r = c_r·s_r` (`c_r > 0`) produces the identical plan.
+The seeded property test (`tool-dag-ecraf-normalization.test.ts`) replays the same
+batch under independent power-of-two memory/CPU rescaling, both bounded and
+unbounded, with conflicts, held usage, and slot caps, and additionally asserts
+partition completeness, slot bounds, feasibility, conflict invariants, and input
+immutability. The spec's worked example holds: memory scale 4 GiB / CPU scale 8
+gives A = 0.625 < B = 0.75 in both GiB and byte units.
+
+Unit normalization is **not** a fairness guarantee or an optimality proof, and it
+is not equivalent to DRF.
+
+## What this is not
+
+- **No live path calls the planner.** `runDagFrontier()` in
+  `packages/agent/src/agent-loop.ts` admits ready calls by source order and
+  settled-claim conflicts; it does not import `tool-dag-ecraf`. There is no
+  shadow recording, feature flag, or default change, and **no measured benefit**
+  is claimed anywhere.
+- **The conflict predicate covers only this pass.** A live caller must re-check
+  unsettled claims, re-validate post-hook arguments, and refuse stale plans
+  before granting (spec §13.4). None of that wiring exists.
+- Starvation/fairness accounting (spec §13.5) and same-budget comparisons
+  (§13.6 steps 4–6) are separate, unstarted work.
+
+## Tests
+
+- `packages/agent/test/tool-dag-ecraf-normalization.test.ts` — version policy,
+  scale derivation, zero-capacity gating, slot cost, overflow/validation
+  boundaries, unit-invariance property test (seed 110917, 300 runs).
+- `packages/agent/test/tool-dag-ecraf.test.ts` — legacy behavior, input
+  rejection, admission invariants, and the §13.3 GiB/bytes ranking-flip example.
+- `packages/agent/test/tool-dag-ecraf-arithmetic.test.ts` — finite-arithmetic
+  overflow rejection, unchanged.
+
+Mutation/negative-control evidence: removing the demand/scale division, weakening
+slot-cost positivity to `< 0`, removing the zero-capacity gate, or removing the
+version whitelist each makes the assertion harness fail (run from a /tmp mutant
+copy; the working tree was not modified during the check).

package/docs/environment-variables.md

@@ -100,7 +100,8 @@ These variables are read by OMK itself. The four built-in harness flags below ar
 | `OMK_YOLO`, `OMK_COMMAND_SAFETY`, `OMK_DISABLE_COMMAND_SAFETY` | Disable the command-safety gate entirely (YOLO mode). `OMK_YOLO` and `OMK_DISABLE_COMMAND_SAFETY` accept `1`, `true`, `yes`, or `on`; `OMK_COMMAND_SAFETY` accepts `0`, `false`, `off`, `disable`, or `disabled`. Every verdict, including block-tier and privilege commands, is skipped in interactive and headless runs. Use only when a verified outer sandbox owns the boundary |
 | `OMK_COMMAND_SAFETY_ASSUME_YES` | `1` or `true` auto-accepts non-privilege confirm-tier commands in interactive **and headless** runs. Privilege confirmation and block-tier commands remain denied. Use only under a trusted outer sandbox when headless auto-accept is intended |
 | `OMK_GROK_HARNESS` | Default-on native `xai` provider dispatch to the `grok-harness` loadout. `0`, `false`, `off`, or `no` disables it |
-| `OMK_DOMAIN_ROUTING` | Set to `1` to enable general prompt-based domain routing. Native xAI harness dispatch does not require it |
+| `OMK_DEVIN_HARNESS` | Default-on `devin` provider dispatch to the `devin-harness` loadout. `0`, `false`, `off`, or `no` disables it; independent from `OMK_GROK_HARNESS` |
+| `OMK_DOMAIN_ROUTING` | Set to `1` to enable general prompt-based domain routing. Native xAI and Devin harness dispatch do not require it |
 | `VISUAL`, `EDITOR` | External editor fallback when `externalEditor` is unset |
 | `HTTP_PROXY`, `HTTPS_PROXY` | Proxy outbound HTTP requests |
 | `OMK_RESOURCE_GOVERNOR` | Resource-governor mode: `off`, `observe` (default), `adaptive`, or `strict`. Feeds `/resource [probe\|policy]` and `omk doctor resources [--json]`; see the resource governor section in [Settings](settings.md) |

package/docs/index.md

@@ -39,6 +39,7 @@ For the full first-run flow, see [Quickstart](quickstart.md).
 - [Providers](providers.md) - subscription and API-key setup for built-in providers.
 - [Provider Resilience](provider-resilience.md) - retry, failover, quota, and safety-stop recovery.
 - [Native xAI Grok](grok-harness.md) - authentication, weekly SuperGrok usage, presets, and thinking tiers.
+- [Devin SWE-2](devin-harness.md) - presets, effort tiers, the 1M-token context budget and lane rule, and the `devin-harness` loadout.
 - [Containerization](containerization.md) - sandbox omk with OpenShell, Gondolin, or Docker.
 - [Settings](settings.md) - global and project settings.
 - [Environment Variables](environment-variables.md) - process configuration, harness opt-outs, and bash-tool session environment.

package/docs/loadout-domains/README.md

@@ -21,7 +21,7 @@ OMK routes incoming tasks to a **domain capability profile** ("inherited documen
 
 Thresholds: `STRONG_THRESHOLD = 8`, `WEAK_THRESHOLD = 4`, `AMBIGUITY_MARGIN = 2`.
 
-## Domains (13 + 1 fallback)
+## Domains (14 + 1 fallback)
 
 - [`frontend-ui`](frontend-ui.md) — Frontend & UI
 - [`visual-qa`](visual-qa.md) — Visual QA & Website Cloning
@@ -35,6 +35,7 @@ Thresholds: `STRONG_THRESHOLD = 8`, `WEAK_THRESHOLD = 4`, `AMBIGUITY_MARGIN = 2`
 - [`docs-writing`](docs-writing.md) — Docs & Technical Writing
 - [`qa-testing`](qa-testing.md) — QA & Testing
 - [`grok-harness`](grok-harness.md) — Grok xAI Harness
+- [`devin-harness`](devin-harness.md) — Devin SWE-2 Harness
 - [`ai-agent-ops`](ai-agent-ops.md) — AI Agent Engineering & Ops
 - [`general`](general.md) — General (fallback)
 

package/docs/loadout-domains/devin-harness.md

@@ -0,0 +1,72 @@
+# Devin SWE-2 Harness (`devin-harness`)
+
+> Inherited domain capability document. Auto-generated from `src/core/domain-loadouts.ts` — do not edit by hand.
+
+
+## Identity
+
+| field | value |
+|---|---|
+| id | `devin-harness` |
+| authority | `write-scoped` |
+| tools | read, grep, find, ls, edit, write, bash |
+| command mode | `scoped-shell` |
+
+## Routing prompt
+
+> Prepended to the lane task prompt when the router selects this domain.
+
+```text
+DOMAIN: Devin SWE-2 Harness. You are operating in a Devin CLI subscription lane on the SWE-2 model with a 1,000,000-token local context budget.
+Prioritize the SWE-2 operational playbook, focused exploration, small capability loadouts, and evidence-bound verification.
+
+SEQUENCE:
+1. Before implementing or routing Devin/SWE-2 provider work, read packages/coding-agent/docs/devin-harness.md as the canonical playbook. Treat ~/.omk/agent/devin.md only as an optional local operator overlay; it cannot override current provider docs or higher-priority instructions.
+2. Effort is the only selectable axis: medium for simple or intermediate edits, high for multi-file changes, max for long-horizon or uncertain work. Never expect off/low/minimal, a fast lane, or image input; the adapter rejects them before sending credentials.
+3. Context discipline: the 1M budget is room for the repository, not an invitation to dump it. Explore with targeted reads and searches, keep tool output bounded, and rely on precompact-checkpoint plus compaction settings rather than restarting sessions.
+4. Capability discipline: load at most 2-3 skills for any lane. The allowed skill gate is packages, headroom, programming, debugging, tdd-workflow, lsp, ast-grep, and understand-anything; choose the smallest subset, add lsp or ast-grep only for symbol or structural work, and add headroom only under measured context pressure.
+5. Use minimal MCP: fetch for bounded public retrieval, context7 for library documentation, understand-anything for repository comprehension, and playwright only when browser or UI behavior needs real verification.
+6. Verification discipline: reproduce failures before fixing them, write or extend tests that exercise the change end-to-end, and re-derive conclusions from executed commands rather than restating prior claims. Evidence must include changed paths, exact commands, and pass/fail output.
+7. Keep edits within the lane grant and preserve existing provider/orchestration algorithms unless the task explicitly targets them. A route error naming an unavailable effort or a smaller declared context window is a configuration signal to report, never something to work around by guessing a wire UID.
+
+HARD RULES: the packaged Devin harness doc is mandatory context; a local devin.md is optional; medium/high/max are the only efforts; the 1M budget never justifies unbounded dumps; maximum 2-3 active skills; never log the Devin session token, user JWT, or auth.json contents; protect-secrets applies.
+```
+
+## Curated skills (8)
+
+- `packages`
+- `headroom`
+- `programming`
+- `debugging`
+- `tdd-workflow`
+- `lsp`
+- `ast-grep`
+- `understand-anything`
+
+## Curated MCP servers (4)
+
+- `fetch`
+- `context7`
+- `understand-anything`
+- `playwright`
+
+## Curated hooks (6)
+
+- `pre-shell-guard`
+- `protect-secrets`
+- `typecheck-after-edit`
+- `stop-verify`
+- `session-context`
+- `precompact-checkpoint`
+
+## Routing triggers (7)
+
+| kind | pattern | weight |
+|---|---|---|
+| keyword | `devin` | 8 |
+| keyword | `swe-2` | 8 |
+| keyword | `swe2` | 8 |
+| keyword | `cognition` | 6 |
+| keyword | `devin cli` | 8 |
+| keyword | `1m context` | 5 |
+| regex | `\b(devin|swe[- ]?2|cognition)\b` | 7 |

package/docs/metrics.md

@@ -160,7 +160,7 @@ node scripts/tb-mini-suite.mjs --json      # feed a runner
 node scripts/tb-mini-suite.mjs --seed 7    # a different fixed subset
 ```
 
-With identical task metadata, seed, size, and collation, selection is repeatable.
+With identical task metadata, selection version, seed, size, and collation, selection is repeatable.
 The default 15-task subset oversamples easy tasks and prioritizes shorter expert
 time estimates; it is a regression signal, not a population-representative score
 or an agent runtime bound. Selection alone is not a capability result. Scoring
@@ -171,12 +171,35 @@ population. Missing difficulty quotas are filled from unselected tasks using the
 same ordering, so a valid request returns exactly that many distinct tasks.
 `--seed` accepts integers from `0` through `4294967295`. Invalid or missing option
 values and oversized requests exit with code `2`; absent, empty, or non-directory
-task paths exit with code `1`. The existing default subset remains unchanged.
+task paths exit with code `1`.
+
+The JSON output now declares `selectionVersion: 2`. Missing, empty, nonfinite, or
+negative expert-time estimates are `null`, not zero. Within each difficulty band
+and in quota refill, known estimates sort before unknown estimates. A genuine zero
+or fractional estimate remains valid. Difficulty quotas still take precedence, so
+unknown-estimate tasks can be selected to fill a band.
+
+`knownExpertMinutes` sums known estimates; `unknownExpertEstimates` counts selected
+tasks with unknown estimates. `totalExpertMinutes` is `null` when any selected
+estimate is unknown. A nonfinite sum is refused with exit `1`, not serialized as
+an apparently missing total. Expert estimates are not agent timeout limits.
+
+For sizes 1–2, available slots go first to the highest-weight difficulty bands
+(medium, then hard), with normal refill if those bands are unavailable. Task names
+no longer decide which excess band quota is discarded. The normal-size quota rule
+is preserved. Human-readable counts use a `Map`, including for labels such as
+`__proto__` that overlap JavaScript object properties.
+
+This is a selection and output-contract change: default membership can change when
+metadata is incomplete, and the prior default JSON digest is historical only.
+Freeze new task manifests before comparison; do not combine version 1 and 2 runs
+as if their selection policy were identical. The limited flat TOML field reader
+is unchanged; this does not add general TOML syntax support.
 
 Run the offline CLI regression tests without downloading tasks or calling models:
 
 ```bash
-node --test scripts/test/tb-mini-suite.test.mjs
+node --test --test-concurrency=1 scripts/test/tb-mini-suite.test.mjs scripts/test/tb-mini-suite-ranking.test.mjs
 ```
 
 See [the harness roadmap](../../../ROADMAP.md) for the dated OMK versus Terminus-2
@@ -187,8 +210,8 @@ criteria. Planned runtime improvements are not measured benchmark gains.
 
 The checkout-only `scripts/tb21-audit.mjs` audits explicitly selected Harbor jobs
 against a caller-pinned manifest digest. It rejects duplicate tasks/trials, missing
-results or costs, mismatched task checksums or configured model labels, and
-contradictory success records. It never starts a model, picks the latest job, joins
+results or costs, mismatched task checksums or configured model labels,
+missing/invalid completion times, and contradictory success records. It never starts a model, picks the latest job, joins
 requests by timestamp, or rewrites evidence. See [TB 2.1 offline audit](tb21-audit.md)
 for the schema, invocation, error codes, and limitations.
 
@@ -196,14 +219,32 @@ A complete audit means recorded outcomes passed these checks, not that every
 provider request obeyed a single-model contract. Wire provenance, actual billing,
 repeated-trial analysis, and statistical superiority need separate evidence.
 
-### Explicit output-limit validation
-
-For callers that supply `AgentLoopConfig.modelContract`, both the contract's
-`maxOutputTokens` and an explicitly supplied request `maxTokens` must be positive
-safe integers. Invalid explicit values are refused before `provider_request` and
-before calling the provider stream function; they are not treated as absent.
-
-An omitted request limit still leaves provider defaults unchecked by this
-predicate. This change does not activate a contract in the CLI or impose an
-effective cap on compaction and other provider paths. Full run-wide enforcement
-remains a roadmap item.
+### Output-limit validation: availability history
+
+**2026-09-13 source snapshot (`ca75f4e5cc`):** the logical contract, CLI/SDK
+wiring, and final Chat Completions model/output-limit checks are in the committed
+source. [Model dispatch contracts](model-contract.md) defines their coverage;
+[Verified Run](verified-run.md) describes the separate protected execution path.
+Neither is a universal provider billing cap or a new controlled benchmark result.
+See [the roadmap, section 16](../../../ROADMAP.md) for current local release-preparation checks.
+
+The following paragraphs describe the earlier checkout only. Its missing modules
+and failing collection are historical observations, not active release blockers.
+
+**2026-09-08 follow-up:** the current worktree restores the logical contract and
+connects it through the CLI/SDK, including SDK-stream summaries. See
+[Model dispatch contracts](model-contract.md) and ROADMAP §13 for fresh evidence
+and the remaining final-wire/accounting gaps. The warning below records the
+preceding checkout, not the current availability of the restored module.
+
+The previous worktree checkpoint tested positive-safe-integer validation of
+`modelContract.maxOutputTokens` and explicit request `maxTokens`. During the
+2026-09-08 re-verification, the checkout changed: `run-model-contract.ts` and the
+corresponding `AgentLoopConfig.modelContract` surface were absent. The remaining
+`model-contract-output-limit.test.ts` fails collection against that checkout.
+
+Do not treat the historical passing tests as proof that this guard is currently
+available. Restoring or porting the runtime contract requires an explicit source
+baseline decision and fresh send-boundary tests. No missing code was silently
+recreated and no failing test was deleted. Full run-wide enforcement, including
+omitted limits and compaction, remains unverified; see ROADMAP sections 11–12.

package/docs/model-catalog-refresh.md

@@ -3,6 +3,56 @@
 확인일: 2026-09-09. 생성기와 공급자 어댑터를 수정한 뒤 `npm run models:refresh`로
 두 카탈로그를 재생성했다. 생성 파일을 손으로 수정하지 않았다.
 
+## 2026-09-17 후속: OpenCode Go DeepSeek V4.1 ID 변경
+
+[OpenCode Go 공식 endpoint 목록](https://opencode.ai/docs/go/)의 현재 ID는
+`deepseek-v4.1-flash`다. 직접 DeepSeek의 `deepseek-flash`와 구분한다.
+9월 17일 카탈로그 갱신은 새 ID를 반영했지만, 생성기의 V4.1 메타데이터 보정은
+이전 ID만 인식해 `low/max`와 `max_tokens` 설정이 누락됐다.
+
+생성기 조건에 Go의 새 ID를 추가하고 기존 Go ID의 보정도 유지했다. 회귀 테스트는
+공급자별 실제 요청 ID를 사용하며 `off/low/high/max`, 이미지 입력, 출력 상한,
+전송 직전 `thinking`과 `reasoning_effort` 검사를 그대로 유지한다.
+ID만 교체한 상태에서도 5개 실패가 재현됐으며 메타데이터 수정 후 통과했다.
+
+`node packages/ai/scripts/generate-models.ts`로 생성한 결과 중 해당 Go 모델의
+메타데이터 변경만 포함했다. 실시간 목록에서 함께 발생한 다른 모델의 추가/삭제,
+가격/상한 변경은 이번 수정에 포함하지 않았다. 생성 파일 값을 수작업으로 만들지 않았다.
+공급자 추론이나 계정별 사용 가능 여부는 검증하지 않았다. 아래 9월 10일 표는 당시 기록이다.
+
+## 2026-09-17 갱신: OpenRouter Union Alpha (stealth)
+
+`stealth/union-alpha` 추가 요청으로 `npm run models:refresh`를 종료0으로 재생성했다.
+생성 파일은 손대지 않았다. 전 소스가 응답했고 `--allow-partial`은 쓰지 않았다.
+
+| 항목 | 값 |
+| --- | --- |
+| 요청 ID | `stealth/union-alpha` (OpenRouter) |
+| context / maxTokens | 262,144 / 131,072 |
+| 입력 | text, image |
+| tool 지원 | `tools`, `tool_choice` |
+| 가격 | prompt/completion 모두 `0` |
+| **thinking** | **없음 — route가 `reasoning`을 선언하지 않음** |
+
+목록과 `/api/v1/models/stealth/union-alpha/endpoints` 모두
+`supported_parameters`가 `max_tokens, temperature, top_p, tools, tool_choice,
+response_format`이다. `reasoning`도 `include_reasoning`도 없다. 선언이 없으므로
+수준을 지어내지 않고 `reasoning: false`로 들어갔으며 `thinkingLevelMap`도 없다.
+이름이 frontier 계열을 연상시킨다는 이유로 effort를 이식하지 않는다.
+
+`created`는 2026-09-16으로 직전 갱신(09-09) 이후에 생긴 항목이다. 한 모델만
+집어넣는 경로가 없어 카탈로그 전체가 8일치 드리프트를 함께 반영한다.
+OpenRouter 371 → 376, 전체 추가 57 · 제거 28(고유 4)이다. 제거는
+`deepseek.r1-v1:0`과 mistral `devstral-small-2` · `mistral-medium` · `pixtral-12b`로,
+갱신 소스에서 더 이상 선정되지 않았다는 뜻이며 공급자의 폐기 공지나 모든 계정의
+사용 불가를 뜻하지 않는다.
+
+가격 `0`은 stealth 공개 기간의 목록 값이다. 무상 사용을 보장하지 않으며 stealth
+해제 시 달라질 수 있다. 실제 청구는 계정에서 따로 확인한다.
+
+표적 검사 `latest-model-thinking` · `catalog-thinking` · `latest-thinking-payload`
+51개가 통과했다. 공급자 추론은 호출하지 않았다.
+
 ## 2026-09-10 재검증: DeepSeek V4.1 Flash 제공 경로
 
 공식 문서·공개 API에서 확인한 기존 OMK 공급자 4곳을 반영했다.
@@ -114,7 +164,7 @@ OpenRouter의 `reasoning.supported_efforts`로 표시할 수준을 만들고,
 
 | 모델·route | 이번 정합성 규칙 |
 | --- | --- |
-| GPT-6 Astra, OpenAI/Responses 및 OpenRouter | low·medium·high·xhigh·max. off/minimal 미노출 |
+| GPT-6 Astra, OpenAI/Responses 및 OpenRouter | low·medium·high·xhigh·max. OMK `ultra`는 공식 천장 `max`의 선택기 별칭이며 와이어에 `ultra`를 보내지 않음. off/minimal 미노출 |
 | Claude Opus 5, Anthropic Messages | adaptive thinking, low·medium·high·xhigh·max. off는 high 이하에서 가능 |
 | Claude Opus 5, Bedrock | legacy budget 대신 adaptive 및 xhigh 사용. application profile의 표시명 매칭 보존 |
 | Gemini 3.7/3.8 Flash, Google/Vertex | low·medium·high. minimal은 API 오류. SDK의 off 요청도 LOW로 처리하며 완전 비활성화로 주장하지 않음 |