mind 5.4.1 — edge-merge follow-up

Follow-up to 5.4.0, caught by a third adversarial verification pass: the _pruned_edges stripping introduced in 5.4.0 ran after the live-edge merge, so it could clobber an edge legitimately (re)created the same session — most visibly the possible-conflict link dream creates when a conflicting pair's old edge decayed the same night. Now stripped on the disk copy before the merge, and _deleted/_pruned_edges are cleared after each save so a stale prune record can't poison a later write. Two regression tests (88 total).

Install pin: raw.githubusercontent.com/Da7-Tech/mind/v5.4.1/mind.py sha256 642fb8e4c445ef4fe6e7739e07547d704015622209cb759094f61d9ece059b35.

mind 5.4.0 — second adversarial audit, hardened

A second adversarial audit (an Opus-4.8 fleet — 17 reviewers with distinct methodologies, each finding independently reproduced-or-refuted, plus a completeness critic) surfaced 8 real defects. All fixed, each with a regression test (86 tests).

Critical (data loss): export_to_agents could silently destroy a user's whole CLAUDE.md/AGENTS.md/GEMINI.md if it merely contained the phrase "mind working memory" — the stale-block heuristic is now a strict structural match.

Security: a symlinked .mind/dreams or .mind/cortex let dream/promote overwrite arbitrary files outside the project — every internal write now rejects symlinked parent directories.

Correctness: a future last_accessed (clock skew on synced memory) inflated node weight unboundedly (now clamped to [0,1]); edge decay was computed but never persisted, so the synaptic-homeostasis claim was false across real CLI runs (now saved — which surfaced and fixed a latent pruned-edge merge-revival bug); corrupt graph.json (non-numeric weight / bad keys) bricked every command (now coerced and validated on load); link created phantom edges by hashing raw text (now shares remember's sanitizer).

Honesty: the soak now also shells out to the real CLI (argv + disk-reload path); the light-sleep phase is documented as the telemetry it is, not a replay.

Install pin: raw.githubusercontent.com/Da7-Tech/mind/v5.4.0/mind.py sha256 50becf1e0f93e79c1fb411ce2aedc663d18bf77931c90cff6cfd3b9c72ce8ee4.

mind 5.3.0 — audit-hardened

Three independent audits, every confirmed finding fixed with a regression test. Highlights: agent-reachable reinforcement (confirm <id> + ids in recall output), dynamic edge weights (synaptic homeostasis — pruning was dead code before), fsync durability, symlink-safe lock/archive with an archive-or-keep guarantee, orphan-edge recovery, merge-on-collision correct, multi-word AR normalization. CI now: SHA-pinned actions, least privilege, Windows in the matrix (9 jobs green). 77 tests. Install pin: raw.githubusercontent.com/Da7-Tech/mind/v5.3.0/mind.py sha256 497a155d…c85c.

mind 5.2.0 — Cursor/Windsurf/Cline/Roo export targets

First release with community code: @Abhinav-0311 added .cursorrules, .windsurfrules, .clinerules and .roo/rules/mind.md export targets (#6) with symlink-parent protection and Windows-portable atomic writes.

Follow-up hardening: tool dotfiles are adopted, not imposed — written only in projects that already use the tool, so a fresh init stays clean; dangling-symlink parents are skipped safely.

66 tests · benchmark unchanged (recall@1 0.95, 0.47 ms) · 180-day soak PASS.

mind 5.1.0 — soak-hardened forgetting

A 180-day simulated-usage soak test (real code, injected clock, now in CI) caught two calibration bugs unit tests couldn't: monthly-cadence facts died one day before their first recall, and decayed weight vetoed exact matches. Fixed with a 45-day grace window, weight-floor ranking, ~2-weeks-per-recall stability, and archive-on-prune (nothing is ever destroyed). Day-180 results: core-fact survival 15/15 (daily/weekly/monthly), stale junk surviving 0/256, graph bounded, 0.37 ms recall on the aged graph. 61 tests.

mind 5.0.0 — first public release

Brain-like memory for any coding agent. One stdlib-only Python file.

Measured (reproduce with python3 bench/bench.py):

• recall@1 0.95, median 0.46 ms @ 100 nodes / 2.2 ms @ 1,000 nodes
• dream determinism: PASS — identical state → identical consolidation plan
• 58 tests, CI green on ubuntu/macos × Python 3.9/3.12/3.14

What's inside

• Spreading-activation recall (RRF + IDF fusion, pattern completion & separation)
• Ebbinghaus forgetting — reinforced memories harden, trivia fades
• Deterministic dream cycle (zero LLM calls, zero tokens, --dry-run, explained journal)
• correct — reconsolidate wrong memories with history
• Export to AGENTS.md / CLAUDE.md / GEMINI.md behind guard markers
• Bilingual EN/AR tokenization; atomic symlink-refusing writes; concurrency-safe saves

Install

curl -O https://raw.githubusercontent.com/Da7-Tech/mind/main/mind.py
python3 mind.py init

Sister project: dream — a sleep cycle for the memory files your agent already has.