Commit Graph

12 Commits

Author SHA1 Message Date
ruv b9010cb226 docs: drop unverified Flash Attention speedup claim from CLAUDE.md (#2739 dream-cycle)
Every dream-cycle for 7+ nights has flagged the "Flash Attention
2.49x-7.47x speedup" number as unverified. It was inherited from
upstream marketing and never reproduced in-tree — same class as the
HNSW "150x-12,500x" claim I dropped in the 2026-07-26 sweep.

Two options were: (a) benchmark it, or (b) drop the number. I'm
taking (b) because a benchmark harness that measures Flash Attention
against a real transformer forward pass is a real project on its own
and would delay closing the credibility flag by weeks.

Change in all 5 sites (root, v3/, v3/@claude-flow/cli/ CLAUDE.md):
- Perf-target rows now read "integration available; measured speedup
  pending benchmark" and explain WHY the number was dropped.
- Sibling bullets under intelligence-system notes cross-reference the
  same explanation so the removal is documented, not silent.

Refs: dream-cycle #2739 (2026-07-20) + repeat mentions in
#2727 #2739 #2752 #2760 #2763 #2768 #2778 #2783.

Co-Authored-By: RuFlo <ruv@ruv.net>
2026-07-26 20:21:07 -04:00
rUv 1d5b06e4a2 fix(federation): unblock main — TS shim types + dispatcher import + witness re-sign (#2604)
* fix(federation): sync shim types + guard optional close() to unblock Build V3

Fixes 7 TS errors on main that block Build V3 across all 3 platforms:
- src/transport/midstream-aware-loader.ts:39-52 — local AgentMessage/
  AgentTransport shim was over-minimized by #2578. Restored required
  {id, type, payload, metadata?} on AgentMessage and made send/onMessage
  2-arg (address, message) to match how plugin.ts actually calls them.
- src/transport/midstream-aware-loader.ts:84 — cast dynamic import()
  through `unknown` since upstream's exported AgentTransport carries a
  richer InboundMessageHandler surface we intentionally hide behind the
  minimal shim.
- src/plugin.ts:481 — this.transport.close is optional (`?`) on the
  shim; guard with optional chaining `close?.()`.

Local tsc build clean after fix.

Co-Authored-By: RuFlo <ruv@ruv.net>

* fix(federation): dispatcher import + re-sign witness manifests (3.25.3 drift)

Two more preexisting main-red root causes, in the same fix:

(1) inbound-dispatcher.ts:22 still imported `AgentMessage` from
    `agentic-flow/transport/loader` — the exact phantom subpath #2578
    eliminated in midstream-aware-loader.ts. Repointed to the local
    shim. This is the residual site #2578 missed; the graph schema smoke
    (ADR-130 P1) failure was a cascade of this same TS type collision.

(2) Re-signed verification/{linux,macos,windows}/manifest.md.json via
    plugins/ruflo-core/scripts/witness/regen.mjs. The 3.25.3 release
    (PR #2602) edited the ADR-104-transport `desc` string in all 3 per-OS
    manifests but did not re-run regen.mjs, so the stored
    integrity.manifestHash no longer matched sha256(JSON.stringify(
    manifest)) and Witness verify exited 1 across all 3 platforms.
    Manifests now verify cleanly (117 pass, 0 drift, 0 regressed).
    Note: witness/regen.mjs takes --manifest + --history and re-signs
    deterministically from manifest.gitCommit; already wired but the
    3.25.3 release process skipped it.

Local verify: pass=117 drift=0 regressed=0 missing=0 on all 3 OSes.
Local tsc build: clean.

Co-Authored-By: RuFlo <ruv@ruv.net>

* fix(hooks): PreToolUse hook emits valid JSON verdict (#2613)

Cursor imports third-party Claude Code hooks under its stricter
`preToolUse` contract, which requires stdout to be a valid JSON
permission verdict (`{"permission":"allow"|"deny"|"ask"}`) and
fail-closes on any other text. The prior PreToolUse commands ran the
ruflo hook shim which prints telemetry to stdout — Cursor then blocked
every Bash/Write/Edit tool call across every workspace.

Fix:
1. ruflo-hook.sh (all three copies) — redirect stdout to /dev/null in
   addition to stderr. Claude Code doesn't consume this stdout either,
   so silencing it is a pure cleanup with no functional cost.
2. hooks.json PreToolUse commands — after invoking the shim, always
   `printf` a well-formed `{"permission":"allow"}` verdict. Valid for
   both Claude Code's JSON-hook contract and Cursor's stricter
   preToolUse contract, so telemetry keeps working while nothing is
   blocked. Also guard `${CLAUDE_PLUGIN_ROOT:-}` so an unset env var
   collapses to the safe empty branch instead of a bash parse error.

Simulated Cursor invocation (CLAUDE_PLUGIN_ROOT unset) now emits:
  {"permission":"allow"}

Co-Authored-By: RuFlo <ruv@ruv.net>

* fix(ci): indirect optional-dep imports in cli — @ruvector/*, @metaharness/router (#2608)

Same install-safety anti-pattern as #2586's prime-radiant fix, now
surfaced in @claude-flow/cli once plugin-agent-federation's TS2307
stopped masking it (630034ca4 on this branch). Build V3 fails with
"Cannot find module '@ruvector/learning-wasm'" and
"Cannot find module '@metaharness/router'" whenever the optional deps
aren't installed — which is the case in the install-safety /
Build V3 configurations.

Every call site is already runtime-guarded (try/catch, availability
probes). But a literal `await import('@ruvector/learning-wasm')` or a
`typeof import('@metaharness/router')` type expression forces tsc to
statically resolve the specifier at build time, so TS2307 fires when
the package is absent.

Fix: route the specifier through a string-typed variable
(`const pkg: string = '...'; await import(pkg)`) — same idiom as the
optional better-sqlite3 loader and #2586. Runtime behaviour unchanged:
the package still loads when present, still falls back cleanly when
absent.

Also replaced services/ruvector-training.ts's static `import type
{ WasmMicroLoRA, ... } from '@ruvector/learning-wasm'` with local
`type X = any` aliases (mirrors the existing FlashAttention et al.
pattern a few lines below).

Verified: with @ruvector/learning-wasm, @ruvector/attention, and
@metaharness/router hidden locally (matching the CI condition), the
cli package now compiles clean (tsc, 0 TS2307). plugin-agent-federation
remains clean as well.

Co-Authored-By: RuFlo <ruv@ruv.net>

* fix(smoke): add metaharness optionalDep to ruflo wrapper (ADR-150 constraint #2)

The all-plugins-smoke, metaharness-ci `score`, and no-metaharness-smoke
CI jobs all fail on the same single step (ruflo-metaharness smoke #16):

  → 16. ruflo wrapper has metaharness in optionalDependencies
        (architectural constraint #2) ... FAIL:
        ruflo wrapper missing metaharness optionalDep

ADR-150 §"Architectural constraint" rule #2 requires `metaharness` and
`@metaharness/*` packages to live in `optionalDependencies` (never
`dependencies`) so ruflo remains operational when they're removed. The
CLI already satisfies this; the wrapper package.json had an empty
`optionalDependencies` block.

Pin to `~0.3.1` — the current npm-published `metaharness@latest`, and
the same range family used by CLAUDE.md's `@metaharness/router@~0.3.2`
pin. Wrapper install remains graceful-fail when the optional dep is
absent (that's the whole point of ADR-150's removable rule #1).

Unblocks PR #2604 merge. Only 3 CI failures on the branch — all
the same rule — so this single change flips them green.

Co-Authored-By: RuFlo <ruv@ruv.net>

* fix(smoke): reconcile ADR-150 rule #2 with #2561 npx-startup budget

Two CI gates were mutually incompatible on this PR:

- ruflo-metaharness smoke #16 (older): asserted \`metaharness\` MUST be
  in \`ruflo/package.json\` optionalDependencies.
- guard #2561 (newer): enforces \`RUFLO_MAX=0\` optionalDependencies
  AND FORBIDs \`metaharness\` specifically, because installing it as a
  ruflo postinstall dep timed out the \`npx -y ruflo@alpha --version\`
  fast-path (measured incident).

Adding \`metaharness\` to ruflo optDeps satisfies #16 but breaks #2561;
removing it satisfies #2561 but leaves #16 red (pre-existing main state).

Fix: correct the smoke #16 rule. ADR-150 §"Architectural constraint"
rule #2 says "when \`@metaharness/*\` or \`metaharness\` appears in a
package's dep graph, it MUST be in \`optionalDependencies\`, never
\`dependencies\`". It does NOT require ruflo to depend on metaharness.
The old check was over-restrictive — it demanded PRESENCE, not just
correct SHAPE-when-present. The new check enforces the actual invariant:
no \`metaharness\` / \`@metaharness/*\` in ruflo's hard \`dependencies\`.
Absence is fine (ruflo delegates to @claude-flow/cli, which keeps
metaharness as a runtime dynamic import via ADR-150 rule #1).

#2561 is stronger evidence (proven cold-npx timeout) than the old #16
assertion, so the correction goes to #16.

Verified locally:
  smoke #16 PASS (new logic)
  guard #2561 PASS

Also reverts the ruflo/package.json optionalDeps change that briefly
tried to satisfy the old #16 rule — the wrapper stays clean.

Co-Authored-By: RuFlo <ruv@ruv.net>

* fix(init): heal duplicate MCP registrations without breaking plugin refs (#2612)

Users who followed pre-rename setup docs ended up with two independent
MCP server registrations pointing at the same binary — one keyed
`claude-flow`, one keyed `ruflo` — doubling Claude Code's tool-schema
load (~350 duplicated tools) with zero added capability. Bug #2612.

The naive fix (rename the canonical key from `claude-flow` → `ruflo`)
silently re-opens #2206 (closed), because ~166 plugin files across
plugins/**/{agents,commands,skills}/*.md reference tools with the
`mcp__claude-flow__*` prefix. Claude Code derives the tool prefix from
the MCP server key; flipping the key to `ruflo` renames every plugin
tool to `mcp__ruflo__*` and every one of those 166 refs errors with
"Unknown tool".

Correct reconciliation:

  - Canonical MCP key STAYS `claude-flow` (preserves #2206 for the 166
    plugin refs). The binary invoked is `ruflo@latest` — only the
    registration name is legacy.
  - Duplicate is HEALED via `ruflo doctor`, not by renaming: detect
    coexisting `ruflo`-keyed + `claude-flow`-keyed registrations for
    the same binary, warn, and instruct the operator to remove the
    `ruflo`-keyed one.

Changes:

  - src/init/mcp-generator.ts — server key restored to `claude-flow`;
    manual `claude mcp add …` commands restored to name `claude-flow`.
    Comment now explains the #2206/#2612 duality.
  - src/init/executor.ts — writeMCPConfig skip-comment updated to
    reflect that `claude-flow` is canonical and `ruflo`-keyed is the
    legacy duplicate form; existing detection accepts both keys (was
    already correct).
  - src/init/claudemd-generator.ts — generated CLAUDE.md `Setup`
    section restored to `claude mcp add claude-flow -- …`.
  - src/commands/doctor.ts — inherit the concurrent session's
    duplicate-detection logic (rufloLocations / duplicateLocations /
    legacyLocations / currentLocations), but flip which key is
    treated as canonical: `claude-flow`/`claude-flow_alpha` are
    current, `ruflo`/`ruflo_alpha` are legacy. Fix-message now
    references removing the `ruflo`-keyed entry and cites #2206.
  - src/commands/swarm.ts — help output uses `claude mcp add
    claude-flow` again.
  - __tests__/issue-2612-mcp-rename.test.ts — regression guard for the
    duplicate-detection warning; header comment documents the
    canonical/legacy roles matching the flipped semantics.
  - CLAUDE.md / v3/@claude-flow/cli/CLAUDE.md / v3/@claude-flow/mcp/CLAUDE.md /
    v3/implementation/init/API.md — kept the binary-update from
    `@claude-flow/cli@latest` → `ruflo@latest` (correct, this IS the
    post-rename wrapper), but restored the MCP server KEY to
    `claude-flow` in every example line.

Verified:
  - tsc clean (CLI package)
  - all 70 test files / 726 tests pass locally, including:
    - __tests__/init-wizard-bugs.test.ts (#2206 canonical key + #2207 dual-key detection)
    - __tests__/stale-mcp-key-2369.test.ts (10 tests, unchanged)
    - __tests__/issue-2612-mcp-rename.test.ts (new — asserts doctor
      surfaces "Duplicate Ruflo MCP registrations found" with legacy
      `ruflo` + current `claude-flow`)
  - Manual doctor run against a synthetic .mcp.json/.claude.json pair
    surfaces the exact warning + remediation string.

Co-Authored-By: RuFlo <ruv@ruv.net>
2026-07-09 10:16:55 -04:00
ruv 0988d92ce8 feat(intelligence): deterministic Tier-1 codemods replacing Agent Booster (ADR-143)
The ADR-026 Tier-1 "Agent Booster, $0, no LLM" path was dead code AND wired
wrong (it fed natural-language instructions to a snippet-merge engine that
expects target code), and the module didn't even resolve. Replace it with real
deterministic codemods so the $0/no-LLM claim is literally true.

- engine.ts: TS-compiler codemods (var-to-const, remove-console, add-logging)
  with formatting-preserving text-range edits + a parse-diagnostic safety net.
- scope-analysis.ts: function-scope reassignment resolution (var→const promotes
  more vars to const without ever producing an incorrect const).
- enhanced-model-router: Tier-1 gated on deterministic intents; route-time
  dry-run only emits [CODEMOD_AVAILABLE] when the edit actually changes a file;
  add-types/add-error-handling/async-await honestly route to a model.
- hooks_codemod MCP tool: real executable Tier-1 surface — single file, files[],
  or glob batch; $0; path-contained.
- Honest [CODEMOD_AVAILABLE] emitters replace [AGENT_BOOSTER_AVAILABLE] and the
  phantom agent_booster_edit_file reference; CLAUDE.md Tier-1 tables corrected.
- Benchmark + golden corpus (12 cases, 100% correct, ~0.55ms/edit, $0) with a
  CI guardrail test; run JSON in cost-trend schema.

Tests: codemod-engine (23), codemod-routing (4), codemod-corpus (13); full CLI
suite 2086 pass / 46 skipped. Closes #2238.

Co-Authored-By: RuFlo <ruv@ruv.net>
2026-05-29 15:40:11 -04:00
rUv 08b3623759 fix+perf+docs(intelligence): self-learning audit, hardening fixes, honest perf numbers (3.10.7) (#2228)
* docs(review): empirical capability audit of the intelligence/self-learning system

6 parallel auditors ran real measurements against built dist (not docs).
- Self-learning loop is REAL, measured end-to-end (pattern confidence +
  Q-routing change persist cross-process; MoE gating converges; SONA WASM
  adapt 0.0042ms; Int8 3.92x; RaBitQ 32x memory — all CONFIRMED).
- Performance multipliers largely UNSUBSTANTIATED: HNSW 150x-12,500x measured
  1.48x peak; Flash Attention 2.49-7.47x fabricated via Math.random() at
  runtime; embeddings 75x / RaBitQ 2.70x retrieval never benchmarked.
- 1 critical bug: CLI inverts negative reward (route feedback -r -1.0 -> +1.00).
- Real-but-inert: WASM MicroLoRA apply(), MCP consolidation gradient (synthetic),
  hooks_intelligence_learn (cosmetic), CLI trajectory-* (no-op stubs), silent
  mock-embedding mislabel.

No source modified — audit branch only. Prioritized fix list in the report.

Co-Authored-By: RuFlo <ruv@ruv.net>

* fix+perf(intelligence): harden self-learning system + replace fabricated/unsubstantiated perf numbers with measured ones

Implements the prioritized fix list from docs/reviews/intelligence-system-audit-2026-05-29.md.

FIXES
- CRITICAL (audit #1, #2222 follow-up): CLI inverted negative reward — route feedback
  -r -1.0 / --reward -1.0 parsed as +1.00, so negative feedback REINFORCED the bad
  agent. Fixed in src/parser.ts (isFlagValue accepts negative numeric literals). All
  three syntaxes now yield -1.0; +6 regression assertions; 52/52 parser tests pass.
- audit #2: removed fabricated Flash Attention speedup (randomized/midpoint telemetry)
  in attention-coordinator (swarm + integration) -> unmeasured sentinel + unverified.
- audit #3: generateEmbedding returns backend:onnx|mock, surfaced in bridge_status /
  import_claude so mock embeddings are never mislabeled as real ONNX.
- audit #4/#5: trajectory-end no longer feeds EWC a synthetic sine gradient;
  hooks_intelligence_learn runs a real distill/consolidate cycle. New test added.

OPTIMIZE (HNSW, measured): root cause — HNSW never ran (no storagePath -> native DB
lock held by daemon -> silent catch{} -> brute force). Fixed vector-db.ts (unique
storagePath, hnswConfig m:32/efC:200, visible fallback warning). Same-harness
before->after: N=5000 0.92x->3.2-4.7x, N=20000 0.95x->1.89x; N=1000 below crossover
(reported honestly); reverted m16/efC64 (recall loss).

BENCHMARK: new scripts/benchmark-intelligence.mjs (HNSW N=1k-50k, Int8 3.84x/cos
0.99999, RaBitQ 32x/0.60ms, SONA 0.0043ms, MoE 0.13->0.88, embedding backend).

DOCS: README + 3x CLAUDE.md + READMEs perf tables now show MEASURED values or
unverified/target; HNSW 150x-12,500x & Flash 2.49-7.47x marked NOT reproduced.

Verification: cli build clean (tsc); 161/161 targeted tests pass.

Co-Authored-By: RuFlo <ruv@ruv.net>

* chore(release): bump 3.10.6 → 3.10.7 (intelligence hardening + honest perf docs)

Co-Authored-By: RuFlo <ruv@ruv.net>
2026-05-29 12:13:53 -04:00
rUv 66640919cd Checkpoint: File edits
Automatic checkpoint created by Claude Code
- Branch: main
- Timestamp: 2026-01-16T19:21:52Z
- Changes: 8 file(s)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-16 19:21:52 +00:00
rUv d9a558f9bf Checkpoint: File edits
Automatic checkpoint created by Claude Code
- Branch: v3
- Timestamp: 2026-01-14T20:34:47Z
- Changes: 1 file(s)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-14 20:34:47 +00:00
rUv f1b4868201 Checkpoint: File edits
Automatic checkpoint created by Claude Code
- Branch: v3
- Timestamp: 2026-01-13T00:25:35Z
- Changes: 1 file(s)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-13 00:25:35 +00:00
rUv 8da8abbcb8 Checkpoint: File edits
Automatic checkpoint created by Claude Code
- Branch: v3
- Timestamp: 2026-01-13T00:25:17Z
- Changes: 6 file(s)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-13 00:25:17 +00:00
rUv 1eb585d8b6 Checkpoint: File edits
Automatic checkpoint created by Claude Code
- Branch: v3
- Timestamp: 2026-01-09T16:40:13Z
- Changes: 12 file(s)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-09 16:40:14 +00:00
rUv 3b91cdc411 Checkpoint: File edits
Automatic checkpoint created by Claude Code
- Branch: v3
- Timestamp: 2026-01-08T20:53:17Z
- Changes: 12 file(s)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-08 20:53:17 +00:00
rUv 315bdf81b8 Checkpoint: File edits
Automatic checkpoint created by Claude Code
- Branch: v3
- Timestamp: 2026-01-08T03:43:49Z
- Changes: 9 file(s)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-08 03:43:49 +00:00
rUv 57daafff6e Checkpoint: File edits
Automatic checkpoint created by Claude Code
- Branch: v3
- Timestamp: 2026-01-08T00:33:18Z
- Changes: 17 file(s)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-08 00:33:18 +00:00