npm - @zhixuan92/multi-model-agent - Versions diffs - 3.10.3 → 3.10.5 - Mend

@zhixuan92/multi-model-agent 3.10.3 → 3.10.5

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 (15) hide show

package/README.md +2 -2
package/dist/skills/mma-audit/SKILL.md +37 -21
package/dist/skills/mma-clarifications/SKILL.md +1 -1
package/dist/skills/mma-context-blocks/SKILL.md +1 -1
package/dist/skills/mma-debug/SKILL.md +37 -21
package/dist/skills/mma-delegate/SKILL.md +1 -1
package/dist/skills/mma-execute-plan/SKILL.md +1 -1
package/dist/skills/mma-investigate/SKILL.md +37 -21
package/dist/skills/mma-retry/SKILL.md +1 -1
package/dist/skills/mma-review/SKILL.md +37 -21
package/dist/skills/mma-verify/SKILL.md +37 -21
package/dist/skills/multi-model-agent/SKILL.md +1 -1
package/dist/telemetry/recorder.js +28 -13
package/dist/telemetry/recorder.js.map +1 -1
package/package.json +2 -2

package/README.md CHANGED Viewed

@@ -84,7 +84,7 @@ Two ways — pick one:
 ```bash
 mmagent serve                          # 127.0.0.1:7337 by default
-curl -s http://localhost:7337/health   # → {"ok":true,"version":"3.10.3",...}
+curl -s http://localhost:7337/health   # → {"ok":true,"version":"3.10.4",...}
 ```
 For an always-on background install (survives reboots): [launchd / systemd templates](./scripts/README.md).
@@ -287,7 +287,7 @@ Full design rationale: [DIRECTION.md](https://github.com/zhixuan312/multi-model-
 ## What's new
-Latest: **3.10.3** — critical telemetry hotfix + UX polish. **3.10.2 silently dropped every emitted event** because the new emit-time R4 validation tripped on a 1ms clock-skew between `runResult.durationMs` and per-stage `durationMs`. Fixed by enforcing R4 by construction in the event builder. Also: skill manifest backfill on upgrade (resolves the "mma-investigate not registered" bug for users who upgraded across v3.4.0); live-elapsed polling headlines (no more stale repeated elapsed between heartbeats); tiered server stdout (default-quiet, `--verbose` for full firehose with throttled heartbeats). Full history: [CHANGELOG](https://github.com/zhixuan312/multi-model-agent/blob/master/CHANGELOG.md).
+Latest: **3.10.4** — review stages were recording the implementer's model (V3 R3 violation root cause). Now record the actual reviewer's resolved tier + model. Plus: telemetry validation is fully warn-only — events never drop, and cross-field warnings now include actual offending values (model, tokens, totals) so config issues vs lifecycle bugs are distinguishable at a glance. Full history: [CHANGELOG](https://github.com/zhixuan312/multi-model-agent/blob/master/CHANGELOG.md).
 ## Full documentation

package/dist/skills/mma-audit/SKILL.md CHANGED Viewed

@@ -8,7 +8,7 @@ when_to_use: >-
   User asks for a doc/spec/config audit OR a methodology skill
   (superpowers:dispatching-parallel-agents, /security-review) points at one AND
   mmagent is running. Audit on PROSE/SPEC docs — use mma-review for source code.
-version: 3.10.3
+version: 3.10.5
 ---
 # mma-audit
@@ -72,26 +72,42 @@ BATCH_ID=$(echo "$BATCH" | jq -r '.batchId')
 @include _shared/response-shape.md
-## Reading the review verdicts (annotation model — 3.8.1+)
-The terminal envelope includes:
-- `specReviewVerdict: 'not_applicable'` — read-only routes have no spec review stage.
-- `qualityReviewVerdict` — outcome of the single annotation pass.
-- `roundsUsed` — `1` when reviewer ran (annotated or errored), `0` when reviewer was skipped.
-There is no rework loop. The reviewer annotates each finding in place and exits — never gates, never causes the worker to re-run.
-Action per `qualityReviewVerdict`:
-- `'annotated'` — every finding in `findings[]` has `reviewerConfidence` (integer 0-100) and possibly `reviewerSeverity`. Sort or filter by confidence; treat low-confidence findings with skepticism.
-- `'skipped'` — kill switch (`MMAGENT_READ_ONLY_REVIEW=disabled` or per-route `MMAGENT_READ_ONLY_REVIEW_AUDIT=disabled`) bypassed the reviewer. Findings carry no reviewer fields; treat as raw worker output.
-- `'error'` — reviewer call or response parsing failed. Findings have no reviewer fields; fall back to caution.
-### Per-finding reviewer fields
-Every finding the worker emits has the standard fields (`id`, `severity`, `claim`, `evidence`, `suggestion?`). After a successful annotation pass, two more fields are added:
-- `reviewerConfidence` (integer 0-100): how confident the reviewer is that the finding is correct, on-brief, and grounded. Use as a filter (`>=70`) or a sort key for triage.
-- `reviewerSeverity?` (`'high' | 'medium' | 'low'`): only present when the reviewer disagrees with the worker's `severity`. Workers tend to inflate severity; use this to dial down. Trust `reviewerSeverity` over `severity` when present.
+## Reading the findings (3.10.5+)
+The terminal envelope's `results[N].annotatedFindings` is a list of structured
+findings the reviewer extracted and scored from the implementer's narrative.
+Every finding has the same shape:
+| Field | Type | Notes |
+|---|---|---|
+| `id` | string | Reviewer-assigned, e.g. `F1`, `F2`. |
+| `severity` | `'critical' \| 'high' \| 'medium' \| 'low'` | 4-tier. |
+| `claim` | string | One-sentence summary. |
+| `evidence` | string ≥20 chars | Quoted from worker output when grounded. |
+| `suggestion?` | string | Optional fix recommendation. |
+| `reviewerConfidence` | `number \| null` | 0–100 from the reviewer; `null` when emitted via deterministic fallback. |
+| `evidenceGrounded` | boolean | True when `evidence` is a verbatim substring of worker output. |
+### Verdict states (`qualityReviewVerdict`)
+- `'annotated'` — every finding is structured. May be reviewer-emitted (with
+  numeric `reviewerConfidence`) or deterministic-fallback (with
+  `reviewerConfidence: null`). The route ALWAYS reaches `'annotated'` unless
+  the reviewer call itself fails transport.
+- `'skipped'` — kill switch (`MMAGENT_READ_ONLY_REVIEW=disabled`).
+- `'error'` — only when the reviewer call fails transport (network / 5xx).
+### Recommended rendering by the main agent
+1. Show ALL findings — never silently drop. Confidence and grounding are
+   soft signals, not gates.
+2. Default sort: severity (critical → low) then `reviewerConfidence` desc
+   (nulls last).
+3. `severity` is the reviewer's authoritative final value — use it directly.
+4. Mark findings with `evidenceGrounded: false` or
+   `reviewerConfidence < 70` as "lower-trust" (collapsed section, lighter
+   color, or `(low confidence)` annotation). User decides what to do.
+5. Severity-tier counts feed the dashboard via V3 `findingsBySeverity`.
 ## Best practices

package/dist/skills/mma-clarifications/SKILL.md CHANGED Viewed

@@ -12,7 +12,7 @@ when_to_use: >-
   `proposedInterpretation` is a hard gate — the batch is paused, not
   informational. The batch will not complete until the caller responds. Treating
   it as advisory is the clarification-as-info anti-pattern (AP5).
-version: 3.10.3
+version: 3.10.5
 ---
 # mma-clarifications

package/dist/skills/mma-context-blocks/SKILL.md CHANGED Viewed

@@ -12,7 +12,7 @@ when_to_use: >-
   Register once here, then pass the ID via `contextBlockIds` on mma-delegate /
   mma-execute-plan / mma-audit / mma-review / mma-verify / mma-debug /
   mma-investigate. Cheaper and faster than inlining the same content N times.
-version: 3.10.3
+version: 3.10.5
 ---
 # mma-context-blocks

package/dist/skills/mma-debug/SKILL.md CHANGED Viewed

@@ -10,7 +10,7 @@ when_to_use: >-
   read files, reproduce, trace — OR a methodology skill
   (superpowers:systematic-debugging) points at the investigation step. Delegate
   the read/reproduce/trace; the main agent stays on the hypothesis and the fix.
-version: 3.10.3
+version: 3.10.5
 ---
 # mma-debug
@@ -78,26 +78,42 @@ BATCH_ID=$(echo "$BATCH" | jq -r '.batchId')
 @include _shared/response-shape.md
-## Reading the review verdicts (annotation model — 3.8.1+)
-The terminal envelope includes:
-- `specReviewVerdict: 'not_applicable'` — read-only routes have no spec review stage.
-- `qualityReviewVerdict` — outcome of the single annotation pass.
-- `roundsUsed` — `1` when reviewer ran (annotated or errored), `0` when reviewer was skipped.
-There is no rework loop. The reviewer annotates each finding in place and exits — never gates, never causes the worker to re-run.
-Action per `qualityReviewVerdict`:
-- `'annotated'` — every finding in `findings[]` has `reviewerConfidence` (integer 0-100) and possibly `reviewerSeverity`. Sort or filter by confidence; treat low-confidence findings with skepticism.
-- `'skipped'` — kill switch (`MMAGENT_READ_ONLY_REVIEW=disabled` or per-route `MMAGENT_READ_ONLY_REVIEW_DEBUG=disabled`) bypassed the reviewer. Findings carry no reviewer fields; treat as raw worker output.
-- `'error'` — reviewer call or response parsing failed. Findings have no reviewer fields; fall back to caution.
-### Per-finding reviewer fields
-Every finding the worker emits has the standard fields (`id`, `severity`, `claim`, `evidence`, `suggestion?`). After a successful annotation pass, two more fields are added:
-- `reviewerConfidence` (integer 0-100): how confident the reviewer is that the finding is correct, on-brief, and grounded. Use as a filter (`>=70`) or a sort key for triage.
-- `reviewerSeverity?` (`'high' | 'medium' | 'low'`): only present when the reviewer disagrees with the worker's `severity`. Workers tend to inflate severity; use this to dial down. Trust `reviewerSeverity` over `severity` when present.
+## Reading the findings (3.10.5+)
+The terminal envelope's `results[N].annotatedFindings` is a list of structured
+findings the reviewer extracted and scored from the implementer's narrative.
+Every finding has the same shape:
+| Field | Type | Notes |
+|---|---|---|
+| `id` | string | Reviewer-assigned, e.g. `F1`, `F2`. |
+| `severity` | `'critical' \| 'high' \| 'medium' \| 'low'` | 4-tier. |
+| `claim` | string | One-sentence summary. |
+| `evidence` | string ≥20 chars | Quoted from worker output when grounded. |
+| `suggestion?` | string | Optional fix recommendation. |
+| `reviewerConfidence` | `number \| null` | 0–100 from the reviewer; `null` when emitted via deterministic fallback. |
+| `evidenceGrounded` | boolean | True when `evidence` is a verbatim substring of worker output. |
+### Verdict states (`qualityReviewVerdict`)
+- `'annotated'` — every finding is structured. May be reviewer-emitted (with
+  numeric `reviewerConfidence`) or deterministic-fallback (with
+  `reviewerConfidence: null`). The route ALWAYS reaches `'annotated'` unless
+  the reviewer call itself fails transport.
+- `'skipped'` — kill switch (`MMAGENT_READ_ONLY_REVIEW=disabled`).
+- `'error'` — only when the reviewer call fails transport (network / 5xx).
+### Recommended rendering by the main agent
+1. Show ALL findings — never silently drop. Confidence and grounding are
+   soft signals, not gates.
+2. Default sort: severity (critical → low) then `reviewerConfidence` desc
+   (nulls last).
+3. `severity` is the reviewer's authoritative final value — use it directly.
+4. Mark findings with `evidenceGrounded: false` or
+   `reviewerConfidence < 70` as "lower-trust" (collapsed section, lighter
+   color, or `(low confidence)` annotation). User decides what to do.
+5. Severity-tier counts feed the dashboard via V3 `findingsBySeverity`.
 ## Best practices

package/dist/skills/mma-delegate/SKILL.md CHANGED Viewed

@@ -11,7 +11,7 @@ when_to_use: >-
   and keep main context free. If a plan file exists → use mma-execute-plan. If
   the task is audit / review / verify / debug / investigate → use the matching
   specialized skill.
-version: 3.10.3
+version: 3.10.5
 ---
 # mma-delegate

package/dist/skills/mma-execute-plan/SKILL.md CHANGED Viewed

@@ -10,7 +10,7 @@ when_to_use: >-
   superpowers:subagent-driven-development / superpowers:executing-plans —
   workers are cheaper and don't pollute main context. Task descriptors must
   match plan headings verbatim.
-version: 3.10.3
+version: 3.10.5
 ---
 # mma-execute-plan

package/dist/skills/mma-investigate/SKILL.md CHANGED Viewed

@@ -12,7 +12,7 @@ when_to_use: >-
   git-history queries. OR you are about to read 3+ files / run any grep in main
   context — that's the inline-labor-leakage anti-pattern (AP2); delegate to this
   skill instead.
-version: 3.10.3
+version: 3.10.5
 ---
 # mma-investigate
@@ -124,26 +124,42 @@ Each task carries an `investigation` field on its per-task report:
 `workerStatus` is one of `done`, `done_with_concerns`, `needs_context`, `blocked`. When `done_with_concerns`, the per-task report carries `incompleteReason` (`turn_cap`, `cost_cap`, `timeout`, or `missing_sections`). When `needs_context`, the worker flagged a `[needs_context]` bullet under `## Unresolved` — re-dispatch with extra context (anchor paths, a context block, or a clarification turn).
-## Reading the review verdicts (annotation model — 3.8.1+)
-The terminal envelope includes:
-- `specReviewVerdict: 'not_applicable'` — read-only routes have no spec review stage.
-- `qualityReviewVerdict` — outcome of the single annotation pass.
-- `roundsUsed` — `1` when reviewer ran (annotated or errored), `0` when reviewer was skipped.
-There is no rework loop. The reviewer annotates each finding in place and exits — never gates, never causes the worker to re-run.
-Action per `qualityReviewVerdict`:
-- `'annotated'` — every finding in `findings[]` has `reviewerConfidence` (integer 0-100) and possibly `reviewerSeverity`. Sort or filter by confidence; treat low-confidence findings with skepticism.
-- `'skipped'` — kill switch (`MMAGENT_READ_ONLY_REVIEW=disabled` or per-route `MMAGENT_READ_ONLY_REVIEW_INVESTIGATE=disabled`) bypassed the reviewer. Findings carry no reviewer fields; treat as raw worker output.
-- `'error'` — reviewer call or response parsing failed. Findings have no reviewer fields; fall back to caution.
-### Per-finding reviewer fields
-Every finding the worker emits has the standard fields (`id`, `severity`, `claim`, `evidence`, `suggestion?`). After a successful annotation pass, two more fields are added:
-- `reviewerConfidence` (integer 0-100): how confident the reviewer is that the finding is correct, on-brief, and grounded. Use as a filter (`>=70`) or a sort key for triage.
-- `reviewerSeverity?` (`'high' | 'medium' | 'low'`): only present when the reviewer disagrees with the worker's `severity`. Workers tend to inflate severity; use this to dial down. Trust `reviewerSeverity` over `severity` when present.
+## Reading the findings (3.10.5+)
+The terminal envelope's `results[N].annotatedFindings` is a list of structured
+findings the reviewer extracted and scored from the implementer's narrative.
+Every finding has the same shape:
+| Field | Type | Notes |
+|---|---|---|
+| `id` | string | Reviewer-assigned, e.g. `F1`, `F2`. |
+| `severity` | `'critical' \| 'high' \| 'medium' \| 'low'` | 4-tier. |
+| `claim` | string | One-sentence summary. |
+| `evidence` | string ≥20 chars | Quoted from worker output when grounded. |
+| `suggestion?` | string | Optional fix recommendation. |
+| `reviewerConfidence` | `number \| null` | 0–100 from the reviewer; `null` when emitted via deterministic fallback. |
+| `evidenceGrounded` | boolean | True when `evidence` is a verbatim substring of worker output. |
+### Verdict states (`qualityReviewVerdict`)
+- `'annotated'` — every finding is structured. May be reviewer-emitted (with
+  numeric `reviewerConfidence`) or deterministic-fallback (with
+  `reviewerConfidence: null`). The route ALWAYS reaches `'annotated'` unless
+  the reviewer call itself fails transport.
+- `'skipped'` — kill switch (`MMAGENT_READ_ONLY_REVIEW=disabled`).
+- `'error'` — only when the reviewer call fails transport (network / 5xx).
+### Recommended rendering by the main agent
+1. Show ALL findings — never silently drop. Confidence and grounding are
+   soft signals, not gates.
+2. Default sort: severity (critical → low) then `reviewerConfidence` desc
+   (nulls last).
+3. `severity` is the reviewer's authoritative final value — use it directly.
+4. Mark findings with `evidenceGrounded: false` or
+   `reviewerConfidence < 70` as "lower-trust" (collapsed section, lighter
+   color, or `(low confidence)` annotation). User decides what to do.
+5. Severity-tier counts feed the dashboard via V3 `findingsBySeverity`.
 ## Best practices

package/dist/skills/mma-retry/SKILL.md CHANGED Viewed

@@ -10,7 +10,7 @@ when_to_use: >-
   you want to re-try the failed indices only. Prefer this over re-dispatching
   the whole batch or inline-retrying — it's idempotent and preserves the
   original batch's diagnostics.
-version: 3.10.3
+version: 3.10.5
 ---
 # mma-retry

package/dist/skills/mma-review/SKILL.md CHANGED Viewed

@@ -10,7 +10,7 @@ when_to_use: >-
   AND mmagent is running. Delegate so each file reviews on its own worker; the
   main agent only decides what to merge. Review on SOURCE CODE — use mma-audit
   for prose specs / configs.
-version: 3.10.3
+version: 3.10.5
 ---
 # mma-review
@@ -75,26 +75,42 @@ BATCH_ID=$(echo "$BATCH" | jq -r '.batchId')
 @include _shared/response-shape.md
-## Reading the review verdicts (annotation model — 3.8.1+)
-The terminal envelope includes:
-- `specReviewVerdict: 'not_applicable'` — read-only routes have no spec review stage.
-- `qualityReviewVerdict` — outcome of the single annotation pass.
-- `roundsUsed` — `1` when reviewer ran (annotated or errored), `0` when reviewer was skipped.
-There is no rework loop. The reviewer annotates each finding in place and exits — never gates, never causes the worker to re-run.
-Action per `qualityReviewVerdict`:
-- `'annotated'` — every finding in `findings[]` has `reviewerConfidence` (integer 0-100) and possibly `reviewerSeverity`. Sort or filter by confidence; treat low-confidence findings with skepticism.
-- `'skipped'` — kill switch (`MMAGENT_READ_ONLY_REVIEW=disabled` or per-route `MMAGENT_READ_ONLY_REVIEW_REVIEW=disabled`) bypassed the reviewer. Findings carry no reviewer fields; treat as raw worker output.
-- `'error'` — reviewer call or response parsing failed. Findings have no reviewer fields; fall back to caution.
-### Per-finding reviewer fields
-Every finding the worker emits has the standard fields (`id`, `severity`, `claim`, `evidence`, `suggestion?`). After a successful annotation pass, two more fields are added:
-- `reviewerConfidence` (integer 0-100): how confident the reviewer is that the finding is correct, on-brief, and grounded. Use as a filter (`>=70`) or a sort key for triage.
-- `reviewerSeverity?` (`'high' | 'medium' | 'low'`): only present when the reviewer disagrees with the worker's `severity`. Workers tend to inflate severity; use this to dial down. Trust `reviewerSeverity` over `severity` when present.
+## Reading the findings (3.10.5+)
+The terminal envelope's `results[N].annotatedFindings` is a list of structured
+findings the reviewer extracted and scored from the implementer's narrative.
+Every finding has the same shape:
+| Field | Type | Notes |
+|---|---|---|
+| `id` | string | Reviewer-assigned, e.g. `F1`, `F2`. |
+| `severity` | `'critical' \| 'high' \| 'medium' \| 'low'` | 4-tier. |
+| `claim` | string | One-sentence summary. |
+| `evidence` | string ≥20 chars | Quoted from worker output when grounded. |
+| `suggestion?` | string | Optional fix recommendation. |
+| `reviewerConfidence` | `number \| null` | 0–100 from the reviewer; `null` when emitted via deterministic fallback. |
+| `evidenceGrounded` | boolean | True when `evidence` is a verbatim substring of worker output. |
+### Verdict states (`qualityReviewVerdict`)
+- `'annotated'` — every finding is structured. May be reviewer-emitted (with
+  numeric `reviewerConfidence`) or deterministic-fallback (with
+  `reviewerConfidence: null`). The route ALWAYS reaches `'annotated'` unless
+  the reviewer call itself fails transport.
+- `'skipped'` — kill switch (`MMAGENT_READ_ONLY_REVIEW=disabled`).
+- `'error'` — only when the reviewer call fails transport (network / 5xx).
+### Recommended rendering by the main agent
+1. Show ALL findings — never silently drop. Confidence and grounding are
+   soft signals, not gates.
+2. Default sort: severity (critical → low) then `reviewerConfidence` desc
+   (nulls last).
+3. `severity` is the reviewer's authoritative final value — use it directly.
+4. Mark findings with `evidenceGrounded: false` or
+   `reviewerConfidence < 70` as "lower-trust" (collapsed section, lighter
+   color, or `(low confidence)` annotation). User decides what to do.
+5. Severity-tier counts feed the dashboard via V3 `findingsBySeverity`.
 ## Best practices

package/dist/skills/mma-verify/SKILL.md CHANGED Viewed

@@ -10,7 +10,7 @@ when_to_use: >-
   against implemented work BEFORE claiming success. Delegate so each checklist
   item gets independent evidence-gathering on a worker. Use this BEFORE saying
   "done" — never after.
-version: 3.10.3
+version: 3.10.5
 ---
 # mma-verify
@@ -76,26 +76,42 @@ BATCH_ID=$(echo "$BATCH" | jq -r '.batchId')
 @include _shared/response-shape.md
-## Reading the review verdicts (annotation model — 3.8.1+)
-The terminal envelope includes:
-- `specReviewVerdict: 'not_applicable'` — read-only routes have no spec review stage.
-- `qualityReviewVerdict` — outcome of the single annotation pass.
-- `roundsUsed` — `1` when reviewer ran (annotated or errored), `0` when reviewer was skipped.
-There is no rework loop. The reviewer annotates each finding in place and exits — never gates, never causes the worker to re-run.
-Action per `qualityReviewVerdict`:
-- `'annotated'` — every finding in `findings[]` has `reviewerConfidence` (integer 0-100) and possibly `reviewerSeverity`. Sort or filter by confidence; treat low-confidence findings with skepticism.
-- `'skipped'` — kill switch (`MMAGENT_READ_ONLY_REVIEW=disabled` or per-route `MMAGENT_READ_ONLY_REVIEW_VERIFY=disabled`) bypassed the reviewer. Findings carry no reviewer fields; treat as raw worker output.
-- `'error'` — reviewer call or response parsing failed. Findings have no reviewer fields; fall back to caution.
-### Per-finding reviewer fields
-Every finding the worker emits has the standard fields (`id`, `severity`, `claim`, `evidence`, `suggestion?`). After a successful annotation pass, two more fields are added:
-- `reviewerConfidence` (integer 0-100): how confident the reviewer is that the finding is correct, on-brief, and grounded. Use as a filter (`>=70`) or a sort key for triage.
-- `reviewerSeverity?` (`'high' | 'medium' | 'low'`): only present when the reviewer disagrees with the worker's `severity`. Workers tend to inflate severity; use this to dial down. Trust `reviewerSeverity` over `severity` when present.
+## Reading the findings (3.10.5+)
+The terminal envelope's `results[N].annotatedFindings` is a list of structured
+findings the reviewer extracted and scored from the implementer's narrative.
+Every finding has the same shape:
+| Field | Type | Notes |
+|---|---|---|
+| `id` | string | Reviewer-assigned, e.g. `F1`, `F2`. |
+| `severity` | `'critical' \| 'high' \| 'medium' \| 'low'` | 4-tier. |
+| `claim` | string | One-sentence summary. |
+| `evidence` | string ≥20 chars | Quoted from worker output when grounded. |
+| `suggestion?` | string | Optional fix recommendation. |
+| `reviewerConfidence` | `number \| null` | 0–100 from the reviewer; `null` when emitted via deterministic fallback. |
+| `evidenceGrounded` | boolean | True when `evidence` is a verbatim substring of worker output. |
+### Verdict states (`qualityReviewVerdict`)
+- `'annotated'` — every finding is structured. May be reviewer-emitted (with
+  numeric `reviewerConfidence`) or deterministic-fallback (with
+  `reviewerConfidence: null`). The route ALWAYS reaches `'annotated'` unless
+  the reviewer call itself fails transport.
+- `'skipped'` — kill switch (`MMAGENT_READ_ONLY_REVIEW=disabled`).
+- `'error'` — only when the reviewer call fails transport (network / 5xx).
+### Recommended rendering by the main agent
+1. Show ALL findings — never silently drop. Confidence and grounding are
+   soft signals, not gates.
+2. Default sort: severity (critical → low) then `reviewerConfidence` desc
+   (nulls last).
+3. `severity` is the reviewer's authoritative final value — use it directly.
+4. Mark findings with `evidenceGrounded: false` or
+   `reviewerConfidence < 70` as "lower-trust" (collapsed section, lighter
+   color, or `(low confidence)` annotation). User decides what to do.
+5. Severity-tier counts feed the dashboard via V3 `findingsBySeverity`.
 ## Best practices

package/dist/skills/multi-model-agent/SKILL.md CHANGED Viewed

@@ -11,7 +11,7 @@ when_to_use: >-
   tasks — AND mmagent is running. Read this once, pick the matching mma-* skill,
   and delegate there. Applies equally whether the user invoked a superpowers
   methodology skill or asked directly.
-version: 3.10.3
+version: 3.10.5
 ---
 # multi-model-agent (router)

package/dist/telemetry/recorder.js CHANGED Viewed

@@ -67,29 +67,44 @@ function _buildRecorder(opts) {
                 if (!d.enabled)
                     return;
                 const event = buildTaskCompletedEvent(ctx);
-                // Validate against the BASE schema (caps, types, enums) — these are
-                // non-negotiable wire-format constraints. Drop only on those failures.
-                // SuperRefine cross-field rules (R1–R11) are checked separately and
-                // logged as warnings, but do NOT drop the event: backend uses
-                // passthrough so degenerate rows can be filtered at query time, and
-                // 3.10.2's drop-on-superRefine behaviour silently lost real telemetry
-                // on 1ms clock-skew (R4) and other measurement-precision edge cases.
+                // Validation is INFORMATIONAL ONLY — never block emit. Backend uses
+                // passthrough so it stores everything; if mma drops events here, that
+                // data is gone forever and the user has no visibility into what was
+                // suppressed. 3.10.2's drop-on-fail design hid real telemetry from
+                // both operator and dashboard.
                 const baseParsed = TaskCompletedEventSchema.safeParse(event);
                 if (!baseParsed.success) {
-                    console.warn('mma-telemetry: dropping schema-invalid event', {
+                    console.warn('mma-telemetry: schema warning (event still emitted)', {
                         eventId: event.eventId,
                         issues: baseParsed.error.issues.map((e) => ({ path: e.path.join('.'), message: e.message })),
                     });
-                    return;
                 }
-                const refined = ValidatedTaskCompletedEventSchema.safeParse(baseParsed.data);
+                const refined = ValidatedTaskCompletedEventSchema.safeParse(event);
                 if (!refined.success) {
-                    console.warn('mma-telemetry: cross-field validation warning (event still emitted)', {
+                    // Surface the actual offending values alongside the rule name so the
+                    // operator can tell at a glance whether the cause is config (wrong
+                    // values) or code (lifecycle bug). Tag the most informative
+                    // top-level fields plus per-stage models for the common R3/R5/R6
+                    // cross-field cases.
+                    const stageModelsByName = (event.stages ?? []).reduce((acc, s) => {
+                        if (s.name && s.model)
+                            acc[s.name] = s.model;
+                        return acc;
+                    }, {});
+                    console.warn('mma-telemetry: cross-field warning (event still emitted)', {
                         eventId: event.eventId,
-                        issues: refined.error.issues.map((e) => ({ path: e.path.join('.'), message: e.message })),
+                        implementerModel: event.implementerModel,
+                        stageModels: stageModelsByName,
+                        totalDurationMs: event.totalDurationMs,
+                        inputTokens: event.inputTokens,
+                        outputTokens: event.outputTokens,
+                        issues: refined.error.issues.map((e) => ({
+                            rule: e.message,
+                            path: e.path.join('.'),
+                        })),
                     });
                 }
-                enqueue(baseParsed.data);
+                enqueue(event);
             }
             catch {
                 dropped++;

package/dist/telemetry/recorder.js.map CHANGED Viewed

	@@ -1 +1 @@
1	- {"version":3,"file":"recorder.js","sourceRoot":"","sources":["../../src/telemetry/recorder.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACjD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,MAAM,EAAE,MAAM,cAAc,CAAC;AACtC,OAAO,EAAE,mBAAmB,EAAE,MAAM,eAAe,CAAC;AACpD,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAClD,OAAO,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,EAAE,KAAK,EAAE,MAAM,YAAY,CAAC;AACnC,OAAO,EAAE,cAAc,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACjE,OAAO,EAAE,cAAc,EAAE,wBAAwB,EAAE,iCAAiC,EAAE,MAAM,mDAAmD,CAAC;AAChJ,OAAO,EACL,uBAAuB,GAExB,MAAM,2DAA2D,CAAC;AASnE,IAAI,SAAS,GAAoB,IAAI,CAAC;AAEtC,MAAM,UAAU,WAAW;IACzB,IAAI,CAAC,SAAS,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,sDAAsD,CAAC,CAAC;IAC1E,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,CAAW;IAC5C,SAAS,GAAG,CAAC,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,IAAiD;IAC9E,MAAM,QAAQ,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;IACtC,SAAS,GAAG,QAAQ,CAAC;IACrB,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,SAAS,cAAc,CAAC,IAAiD;IACvE,MAAM,EAAE,OAAO,EAAE,cAAc,EAAE,GAAG,IAAI,CAAC;IACzC,MAAM,KAAK,GAAG,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC;IACjC,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,IAAI,UAAU,GAAkB,IAAI,CAAC;IACrC,IAAI,OAAO,GAAG,CAAC,CAAC;IAEhB,MAAM,gBAAgB,GAAG,GAAW,EAAE;QACpC,IAAI,CAAC,UAAU,EAAE,CAAC;YAChB,UAAU,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC,SAAS,CAAC;QACtD,CAAC;QACD,OAAO,UAAU,CAAC;IACpB,CAAC,CAAC;IAEF,MAAM,OAAO,GAAG,CAAC,KAA8B,EAAQ,EAAE;QACvD,IAAI,CAAC;YACH,MAAM,EAAE,GAAG,gBAAgB,EAAE,CAAC;YAC9B,MAAM,IAAI,GAAG,gBAAgB,CAAC,EAAE,SAAS,EAAE,EAAE,EAAE,cAAc,EAAE,CAAC,CAAC;YACjE,MAAM,GAAG,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;YAEpC,KAAK,CAAC,MAAM,CAAC;gBACX,aAAa,EAAE,cAAc;gBAC7B,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,cAAc,EAAE,IAAI,CAAC,cAAc;gBACnC,EAAE,EAAE,IAAI,CAAC,EAAE;gBACX,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,UAAU,EAAE,GAAG;gBACf,MAAM,EAAE,CAAC,KAAK,CAAC;aAChB,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE;gBACZ,OAAO,EAAE,CAAC;YACZ,CAAC,CAAC,CAAC;QACL,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,EAAE,CAAC;QACZ,CAAC;IACH,CAAC,CAAC;IAEF,OAAO;QACL,IAAI,MAAM;YACR,OAAO,UAAU,CAAC,MAAM,CAAC;QAC3B,CAAC;QAED,OAAO;QAEP,mBAAmB,CAAC,GAAG;YACrB,IAAI,CAAC;gBACH,MAAM,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC;gBAC1B,IAAI,CAAC,CAAC,CAAC,OAAO;oBAAE,OAAO;gBACvB,MAAM,KAAK,GAAG,uBAAuB,CAAC,GAAG,CAAC,CAAC;gBAC3C,oEAAoE;gBACpE,~~uEAAuE~~;~~gBACvE~~,oEAAoE;gBACpE,~~8DAA8D~~;~~gBAC9D,oEAAoE~~;~~gBACpE~~,~~sEAAsE;gBACtE,qEAAqE;gBACrE,~~MAAM,UAAU,GAAG,wBAAwB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;gBAC7D,IAAI,CAAC,UAAU,CAAC,OAAO,EAAE,CAAC;oBACxB,OAAO,CAAC,IAAI,CAAC,~~8CAA8C~~,EAAE;~~wBAC3D~~,OAAO,EAAE,KAAK,CAAC,OAAO;wBACtB,MAAM,EAAE,UAAU,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;qBAC7F,CAAC,CAAC;~~oBACH~~,~~OAAO;gBACT,~~CAAC;gBACD,MAAM,OAAO,GAAG,iCAAiC,CAAC,SAAS,CAAC,~~UAAU~~,CAAC,IAAI,CAAC,CAAC;~~gBAC7E~~,IAAI,CAAC,~~OAAO~~,CAAC,OAAO,EAAE,CAAC;~~oBACrB~~,OAAO,CAAC,IAAI,CAAC,~~qEAAqE~~,EAAE;~~wBAClF~~,OAAO,EAAE,KAAK,CAAC,OAAO;wBACtB,MAAM,EAAE,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,~~EAAE,OAAO,EAAE,~~CAAC,CAAC~~,OAAO,EAAE,CAAC,CAAC~~;~~qBAC1F~~,CAAC,CAAC;gBACL,CAAC;gBACD,OAAO,CAAC,~~UAAU~~,CAAC,~~IAA+B,~~CAAC~~,CAAC~~;~~YACtD~~,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,EAAE,CAAC;YACZ,CAAC;QACH,CAAC;QAED,KAAK,CAAC,cAAc,CAAC,OAAO;YAC1B,MAAM,cAAc,CAAC,OAAO,CAAC,CAAC;YAC9B,UAAU,CAAC,KAAK,EAAE,CAAC;YACnB,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,EAAE,wBAAwB,CAAC,CAAC;YAC1D,IAAI,UAAU,CAAC,SAAS,CAAC;gBAAE,UAAU,CAAC,SAAS,CAAC,CAAC;YACjD,UAAU,GAAG,IAAI,CAAC;YAClB,IAAI,OAAO,EAAE,eAAe,EAAE,CAAC;gBAC7B,eAAe,CAAC,OAAO,CAAC,CAAC;YAC3B,CAAC;QACH,CAAC;KACF,CAAC;AACJ,CAAC"}
1	+ {"version":3,"file":"recorder.js","sourceRoot":"","sources":["../../src/telemetry/recorder.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACjD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,MAAM,EAAE,MAAM,cAAc,CAAC;AACtC,OAAO,EAAE,mBAAmB,EAAE,MAAM,eAAe,CAAC;AACpD,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAClD,OAAO,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,EAAE,KAAK,EAAE,MAAM,YAAY,CAAC;AACnC,OAAO,EAAE,cAAc,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACjE,OAAO,EAAE,cAAc,EAAE,wBAAwB,EAAE,iCAAiC,EAAE,MAAM,mDAAmD,CAAC;AAChJ,OAAO,EACL,uBAAuB,GAExB,MAAM,2DAA2D,CAAC;AASnE,IAAI,SAAS,GAAoB,IAAI,CAAC;AAEtC,MAAM,UAAU,WAAW;IACzB,IAAI,CAAC,SAAS,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,sDAAsD,CAAC,CAAC;IAC1E,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,CAAW;IAC5C,SAAS,GAAG,CAAC,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,IAAiD;IAC9E,MAAM,QAAQ,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;IACtC,SAAS,GAAG,QAAQ,CAAC;IACrB,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,SAAS,cAAc,CAAC,IAAiD;IACvE,MAAM,EAAE,OAAO,EAAE,cAAc,EAAE,GAAG,IAAI,CAAC;IACzC,MAAM,KAAK,GAAG,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC;IACjC,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,IAAI,UAAU,GAAkB,IAAI,CAAC;IACrC,IAAI,OAAO,GAAG,CAAC,CAAC;IAEhB,MAAM,gBAAgB,GAAG,GAAW,EAAE;QACpC,IAAI,CAAC,UAAU,EAAE,CAAC;YAChB,UAAU,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC,SAAS,CAAC;QACtD,CAAC;QACD,OAAO,UAAU,CAAC;IACpB,CAAC,CAAC;IAEF,MAAM,OAAO,GAAG,CAAC,KAA8B,EAAQ,EAAE;QACvD,IAAI,CAAC;YACH,MAAM,EAAE,GAAG,gBAAgB,EAAE,CAAC;YAC9B,MAAM,IAAI,GAAG,gBAAgB,CAAC,EAAE,SAAS,EAAE,EAAE,EAAE,cAAc,EAAE,CAAC,CAAC;YACjE,MAAM,GAAG,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;YAEpC,KAAK,CAAC,MAAM,CAAC;gBACX,aAAa,EAAE,cAAc;gBAC7B,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,cAAc,EAAE,IAAI,CAAC,cAAc;gBACnC,EAAE,EAAE,IAAI,CAAC,EAAE;gBACX,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,UAAU,EAAE,GAAG;gBACf,MAAM,EAAE,CAAC,KAAK,CAAC;aAChB,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE;gBACZ,OAAO,EAAE,CAAC;YACZ,CAAC,CAAC,CAAC;QACL,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,EAAE,CAAC;QACZ,CAAC;IACH,CAAC,CAAC;IAEF,OAAO;QACL,IAAI,MAAM;YACR,OAAO,UAAU,CAAC,MAAM,CAAC;QAC3B,CAAC;QAED,OAAO;QAEP,mBAAmB,CAAC,GAAG;YACrB,IAAI,CAAC;gBACH,MAAM,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC;gBAC1B,IAAI,CAAC,CAAC,CAAC,OAAO;oBAAE,OAAO;gBACvB,MAAM,KAAK,GAAG,uBAAuB,CAAC,GAAG,CAAC,CAAC;gBAC3C,oEAAoE;gBACpE,sEAAsE;gBACtE,oEAAoE;gBACpE,mEAAmE;gBACnE,+BAA+B;gBAC/B,MAAM,UAAU,GAAG,wBAAwB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;gBAC7D,IAAI,CAAC,UAAU,CAAC,OAAO,EAAE,CAAC;oBACxB,OAAO,CAAC,IAAI,CAAC,qDAAqD,EAAE;wBAClE,OAAO,EAAE,KAAK,CAAC,OAAO;wBACtB,MAAM,EAAE,UAAU,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;qBAC7F,CAAC,CAAC;gBACL,CAAC;gBACD,MAAM,OAAO,GAAG,iCAAiC,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;gBACnE,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC;oBACrB,qEAAqE;oBACrE,mEAAmE;oBACnE,4DAA4D;oBAC5D,iEAAiE;oBACjE,qBAAqB;oBACrB,MAAM,iBAAiB,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,MAAM,CACnD,CAAC,GAA2B,EAAE,CAAmC,EAAE,EAAE;wBACnE,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,KAAK;4BAAE,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC;wBAC7C,OAAO,GAAG,CAAC;oBACb,CAAC,EACD,EAAE,CACH,CAAC;oBACF,OAAO,CAAC,IAAI,CAAC,0DAA0D,EAAE;wBACvE,OAAO,EAAE,KAAK,CAAC,OAAO;wBACtB,gBAAgB,EAAE,KAAK,CAAC,gBAAgB;wBACxC,WAAW,EAAE,iBAAiB;wBAC9B,eAAe,EAAE,KAAK,CAAC,eAAe;wBACtC,WAAW,EAAE,KAAK,CAAC,WAAW;wBAC9B,YAAY,EAAE,KAAK,CAAC,YAAY;wBAChC,MAAM,EAAE,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;4BACvC,IAAI,EAAE,CAAC,CAAC,OAAO;4BACf,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC;yBACvB,CAAC,CAAC;qBACJ,CAAC,CAAC;gBACL,CAAC;gBACD,OAAO,CAAC,KAA2C,CAAC,CAAC;YACvD,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,EAAE,CAAC;YACZ,CAAC;QACH,CAAC;QAED,KAAK,CAAC,cAAc,CAAC,OAAO;YAC1B,MAAM,cAAc,CAAC,OAAO,CAAC,CAAC;YAC9B,UAAU,CAAC,KAAK,EAAE,CAAC;YACnB,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,EAAE,wBAAwB,CAAC,CAAC;YAC1D,IAAI,UAAU,CAAC,SAAS,CAAC;gBAAE,UAAU,CAAC,SAAS,CAAC,CAAC;YACjD,UAAU,GAAG,IAAI,CAAC;YAClB,IAAI,OAAO,EAAE,eAAe,EAAE,CAAC;gBAC7B,eAAe,CAAC,OAAO,CAAC,CAAC;YAC3B,CAAC;QACH,CAAC;KACF,CAAC;AACJ,CAAC"}

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "@zhixuan92/multi-model-agent",
-  "version": "3.10.3",
+  "version": "3.10.5",
   "type": "module",
   "license": "MIT",
   "description": "Standalone HTTP server for multi-model-agent. Routes tool-invocation work to Claude, Codex, or OpenAI-compatible sub-agents with async-polling REST dispatch and installable skills for Claude Code, Gemini CLI, Codex CLI, and Cursor.",
@@ -52,7 +52,7 @@
   },
   "dependencies": {
     "@asteasolutions/zod-to-openapi": "^8.5.0",
-    "@zhixuan92/multi-model-agent-core": "^3.10.3",
+    "@zhixuan92/multi-model-agent-core": "^3.10.5",
     "gray-matter": "^4.0.3",
     "minimist": "^1.2.8",
     "proper-lockfile": "^4.1.2",