Skip to content

armgit / ruflo

Public Imported from GitHub
0

🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated

dream(security): #3338 record round-2 review response in ledger

Claudecb13478132b4
diff --git a/docs/dream-cycle/LEDGER.md b/docs/dream-cycle/LEDGER.md
index 9c8570a658..f67fff0509 100644
--- a/docs/dream-cycle/LEDGER.md
+++ b/docs/dream-cycle/LEDGER.md
@@ -195,4 +195,4 @@ as a first-class, high-priority candidate, not a background note.
**Update (2026-09-16):** Verified via GitHub MCP tools (`pull_request_read`, not `gh` β€” unavailable in this environment) that the entire backlog flagged above as "unmerged draft"/"awaiting fresh CI" is now **MERGED**: #3086, #3094, #3103, #3110, #3119, #3123, #3130, #3139, #3152, #3160 (the 2026-08-24..09-02 batch) all merged 2026-09-14, and #3184, #3221, #3232, #3243, #3266, #3291, #3302 (2026-09-05..09-12) all merged 2026-09-14/15 β€” 20/20 checked, 0 open. Trailing-night merge rate has gone from "0 of last 14" (as of 08-31) to 20/20 of everything checked tonight. STEP 1.1's "bias toward small, reviewable candidates" guidance is credited but no longer forced by a stuck backlog β€” reviewability remains a goal on its own merits, not a workaround for non-merging.
-| 2026-09-16 | security | Six `hive-mind_*` MCP tools (`spawn`, `consensus` `propose`, `broadcast`, `shutdown`, `memory` `set`/`delete`) mutated hive state with no `hiveToken` check, though sibling `join`/`leave`/`vote` were capability-bound the night before (#3291, merged 09-15) β€” most severely, `spawn` let an unauthenticated caller mint its own voting "workers" and vote as them, bypassing #3291 entirely via a sibling tool; all six now reuse the existing `requireHiveToken` gate | #3338 | #3339 | yes | ACCEPT | stash-isolated: 7/12 new tests fail on baseline exactly for the 6 gated call sites, 18/18 pass on candidate; full `@claude-flow/cli` suite (251 files/3165 tests) failing-test-name sets identical both ways (34 pre-existing, unrelated); `tsc --noEmit` byte-identical (22 errors); independent adversarial critique CONFIRMED-WITH-CAVEATS pre-fix (found the `hive-mind_memory` gap the first patch missed), re-verified after closing it | f2dca5c7... | 20/20 previously-checked backlog PRs now confirmed merged (see note above) β€” first night since 08-31 with zero open dream-cycle backlog |
+| 2026-09-16 | security | Seven `hive-mind_*` MCP tools (`spawn`, `consensus` `propose`, `broadcast`, `shutdown`, `memory` `set`/`delete`, `optimize-memory`) mutated hive state with no `hiveToken` check, though sibling `join`/`leave`/`vote` were capability-bound the night before (#3291, merged 09-15) β€” most severely, `spawn` let an unauthenticated caller mint its own voting "workers" and vote as them, bypassing #3291 entirely via a sibling tool; all seven now reuse the existing `requireHiveToken` gate. **Update (ruvnet review, round 1): REJECTed** β€” correctly identified that gating siblings doesn't "establish authorization" while `hive-mind_init` (the credential-issuance point) remained reachable by any caller to mint/re-obtain the token, that `hive-mind_optimize-memory` was missed, that tests only exercised direct handlers (not real CLI commands), and that the gist's witness didn't reproduce from the committed file alone. All four addressed same night: `hive-mind_init` now requires its own same-machine `bootstrapSecret` (lazily-created 0600 file, same trust boundary `getHiveTokenForCli()` already used) for both bootstrap and re-init and no longer echoes `hiveToken`; `optimize-memory` gated; added real `execFileSync`-spawned `node bin/cli.js hive-mind init/spawn` child-process tests; witness regenerated and independently re-verified to reproduce from the final commit alone | #3338 | #3339 | yes | ACCEPT (round 2, post-REJECT) | round 1: stash-isolated 7/12 new tests fail on baseline for the 6 gated sites; round 2: test file grew to 28 tests (+ bootstrap-secret block + 2 CLI-process tests), 37 total hive-mind tests pass; building previously-unbuilt `@claude-flow/swarm`/`@claude-flow/neural` siblings let `tsc --noEmit` run fully clean (0 errors, was 22 unrelated); full suite 18 pre-existing unrelated failures (down from 34 once those siblings could build), zero referencing hive-mind either round | 4c373db1... (round 2, verified reproducible; round 1's f2dca5c7... did not reproduce β€” disclosed, not hidden) | 20/20 previously-checked backlog PRs now confirmed merged (see note above) β€” first night since 08-31 with zero open dream-cycle backlog |