@mindrian_os/cli 2.0.0-beta.1 → 2.0.0-beta.13
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.
- package/CHANGELOG.md +87 -0
- package/README.md +55 -92
- package/package.json +1 -1
package/CHANGELOG.md
CHANGED
|
@@ -1,3 +1,90 @@
|
|
|
1
|
+
## [2.0.0-beta.13] - 2026-08-28
|
|
2
|
+
|
|
3
|
+
### Added
|
|
4
|
+
- Memory and Context Operator MCP (Phase 270): three new atomic tools -- `context_assemble` (exposes the four-leg `getRoomContext` budget model with `estimate_only` cost preview before paying it), `graph_reason` (transitive support-closure and cross-room nearest-decision queries, both graph-native and previously unreachable by a flat directory walk), and `identity_write` (the first writer `~/.mindrian-user.md` has ever had). MCP Resources fixed to resolve the room per-session instead of once at boot (`room://state`), plus `mos://tree` and `mos://room/{slug}/tree` exposing the ICM forest live. **Honest tradeoff, not a win:** measured tool-schema token budget went **up** ~17% (36->39 tools, ~7,167->~8,377 approx tokens), not down -- the three new atomic tools cost more than the consolidation saved. Recorded as such, not spun.
|
|
5
|
+
- `interactive_first_reward` vocabulary gains a ninth member, `live_deliverable` (Phase 267.3): a real, deployed, shareable artifact from an irreversible action (a live URL, a sent file) -- minted rather than forcing `/mos:publish`'s actual reward into an existing, incorrect term. A `data/first-reward-surfaces.json` registry now lets non-frontmatter surfaces (bash hooks, injected prose) declare a first-reward contract the linter can read; `scripts/session-start`'s four surfaces (FIRST_INSTALL, UPDATE, MODE_ROUTING, COLD_START_MENU) are now declared.
|
|
6
|
+
- `/mos:doctor --check-rs-engine --fix` and direct auto-install: `scripts/compute-hsi.py` and `scripts/rs-engine.py` (the entry points for `/mos:act`, `/mos:mos-reason`, `/mos:find-bottlenecks`) now call the plugin's own already-shipped Python dependency auto-installer on missing imports, instead of hard-failing with a manual `pip install` instruction. Six sibling scripts already had this; these two did not.
|
|
7
|
+
|
|
8
|
+
### Fixed
|
|
9
|
+
- Reference-file citations are now anchored with `${CLAUDE_PLUGIN_ROOT}`, so the files a command actually needs load correctly from a real Data Room instead of only from the plugin dev repo. A bare `references/...` path resolves against your session's current working directory, not the plugin install directory, so these citations worked by pure coincidence in development (where the two happen to be the same folder) and silently failed for every installed user on all three surfaces. **134 of 134 measured sites now anchored** (103 commands/skills/agents/pipelines, plus the final 31 -- 16 methodology commands and `commands/doctor.md` -- unblocked by the `interactive_first_reward` vocabulary fix above and landed clean). Hand-authored skills use the fail-closed `${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?...}}` form, which refuses honestly instead of silently reading the wrong path when a foreign Agent-Skills host loads a skill with no plugin root set. `commands/radar.md`'s five sites are a reasoned exception (a genuine write into a git-tracked source of record, not a read -- anchoring it would redirect writes into the install cache every update wipes) -- allowlisted with a written reason, not silently skipped.
|
|
10
|
+
- Added `scripts/check-plugin-path-anchoring.cjs` to the release gates (`scripts/verify-release`, gate 10c) so this class cannot silently return. A newly authored command, skill, agent, or pipeline file citing a bare plugin-relative path now fails the release gate before it can ship. `data/first-reward-surfaces.json` declarations are checked by a companion gate 10d.
|
|
11
|
+
- Capability Radar Absorption + Routing (Phase 265, 23 plans): a full adversarial code review after all plans passed their own tests caught two real regressions no test suite caught -- `/mos:scout`'s competitor fan-out had Write + Brain access despite promising read-only (fixed by minting a genuinely read-only `agents/competitor-watch-fetcher.md`), and `/mos:vault` had a real human-gate bypass (a file-routing step re-invoked before, not after, the navigator's confirmation).
|
|
12
|
+
- MCP layer correctness fixes (Phase 266): a per-call timeout budget that didn't hold in aggregate (60s vs a 30s host budget), caught by independent re-verification after the phase's own tests passed; fixed and re-measured (15.2s).
|
|
13
|
+
|
|
14
|
+
### Known limitations, stated plainly
|
|
15
|
+
- The other ~50 commands beyond the 17 fixed this release still lack an `interactive_first_reward` declaration (a repo-wide backfill, not yet started; no release gate currently requires full coverage, so this does not block shipping).
|
|
16
|
+
- A code-review pass this cycle found 5 Critical correctness bugs in the local SQLite graph write chokepoint (`lib/core/navigation/edges.cjs`'s `writeEdge` silently discards some writes while reporting success; a schema mismatch fails writes from one of two database openers; the Brain-supplied edge-type allowlist is bypassed on one write path; the busy-timeout fix is applied at 1 of ~20 database openers; a locked room can misreport as having no database at all) -- found and registered as its own phase, **not fixed in this release**. Full detail: `.planning/ROADMAP.md` Phase 273.
|
|
17
|
+
- Phase 134's original objective (eliminating the Python dependency for HSI/reverse-salient analysis entirely) was tracked as complete but was never actually implemented -- the auto-install fix above is a mitigation, not the real fix. Registered as Phase 272, not done here.
|
|
18
|
+
|
|
19
|
+
## [2.0.0-beta.11] - 2026-08-24
|
|
20
|
+
|
|
21
|
+
### Added
|
|
22
|
+
-
|
|
23
|
+
|
|
24
|
+
### Fixed
|
|
25
|
+
- `/mos:organize`'s D-09 soft-alias redirected to `/mos:rooms organize`, a subcommand that never existed -- an unmatched subcommand silently fell through to `list`'s default behavior instead of failing honestly. `commands/rooms.md` now defines `organize` as a real subcommand that reports "not yet implemented" (tracked as `organize-soft-alias-redirects-to-nonexistent-target`) instead of masking the gap, and any other unrecognized `/mos:rooms` subcommand now gets the standard 3-line error instead of a silent `list` fallback. The `room_content` MCP tool's `command` schema also advertised `organize` as a valid enum value with no matching dispatcher case, so any programmatic caller got a runtime `Unknown room_content command: organize` instead of a validation-time rejection; `organize` is removed from that enum until a real handler exists.
|
|
26
|
+
|
|
27
|
+
## [2.0.0-beta.9] - 2026-08-23
|
|
28
|
+
|
|
29
|
+
### Added
|
|
30
|
+
- Roadmap-Type Selector (Phase 264): a navigator's stated research goal is silently classified into one of six scientific-roadmapping output shapes (Landscape Analysis, Technical Roadmap, Pipeline Analysis, Opportunity Analysis, Agenda-Setting Manifesto, Vision Paper) and resolved to the matching framework-name chain via the existing chain_resolve seam. New sensor SENS-18 (`lib/core/sensors/sensor-roadmap-type.cjs`), a deterministic additive-score classifier following `dual-path-detector.cjs`'s template, registered across the 3-array sensor lockstep. New `data/roadmap-type-chains.json` chain table with a five-arm drift validator mirroring the existing `dispatch-framework-map.json` precedent.
|
|
31
|
+
- Challenge-driven execution proof: `find-bottlenecks` (Reverse Salient) now demonstrably runs under the already-shipped `ralph_verify` bounded self-critique seam with a new synchronous, two-pass adversarial critic (`lib/core/salient-governance.cjs`), reusing `eureka-critic.cjs`'s already-ruled 2-judge-unanimous pattern rather than an N-judge panel. Proven via a direct `chainExecutor.runChain()` call (mirroring the existing `debate-composition.cjs` precedent), including a live mutation proof that flipping a gate-condition function makes the test suite go red.
|
|
32
|
+
- `chain-executor.cjs`'s B3/Canon Part 3 stop-condition contract (the chain halts at the first material step, never an autonomous cross-step loop) is now pinned by a scoped sha256 hash over its six gate/stop-condition symbols plus a whole-file zero-diff check, so a future accidental edit there is caught immediately rather than discovered later.
|
|
33
|
+
|
|
34
|
+
### Known limitations
|
|
35
|
+
- The live `chain_resolve`/`chain_run` MCP path does not yet honor `ralph_verify` -- its async execution branch neither calls the retry seam nor passes a `selfCritiqueFn`. The challenge-driven proof above runs via a direct executor call, not the navigator-facing chain flow. Wiring the live path is a named follow-on, not done in this release.
|
|
36
|
+
- The classifier and chain table are proven-composable (integration-tested end to end) but have no wired UI/reach-card surface yet.
|
|
37
|
+
|
|
38
|
+
## [2.0.0-beta.7] - 2026-08-19
|
|
39
|
+
|
|
40
|
+
### Added
|
|
41
|
+
- The dual-graph loop, closed (the "Build the Loop" milestone promise): local sensors supply WHEN, the Brain supplies WHICH and IN WHAT SEQUENCE, a human gate sits between. recommendChain wrapper on the one governed Brain wire (frozen v1 contract, Brain repo docs/RECOMMEND-CHAIN-CONTRACT.md); the brain_framework_chain companion finally has a consumer + chain adapter in chain-recommender (alias normalization, command-map divergence logged as typed memory events, posture authority stays LOCAL); chain offers surface on suggest_next for hookless surfaces (260819-c8j)
|
|
42
|
+
- Hookless-surface runtime protocol served as MCP `instructions` at initialize, plus Desktop prompt-menu entries bind-room / status / act - Desktop and Cowork run the same Larry loop CLI hooks enforce (2026-08-18/19 night session)
|
|
43
|
+
- Doctor class-M layer 6 `store_identity`: resolved endpoint, node-count floor 29,000, the number 28,325 recognized BY NAME as the stale-replica signature, optional GraphRagMeta stamp read (260819-c9b)
|
|
44
|
+
- Dual-graph health gate live reader wired (M4 predicate rewritten to fit the server's 5s read budget, 5031ms to 688ms) and the first conclusive live baseline recorded (260819-c9b)
|
|
45
|
+
- Part 8 in-process egress belt at the callTool chokepoint - free-form payload classification now covers every Brain wrapper, not 4 of 16 (260819-c8j)
|
|
46
|
+
|
|
47
|
+
### Fixed
|
|
48
|
+
- MCP write path enabled on the Claude Code host: graph_write / memory_event / artifact_file (the governed navigation.cjs door) no longer refuse and provoke ungoverned direct-disk filing fallbacks (260819-bql, RCA option A)
|
|
49
|
+
- One resume owner for material_step gates: gate_answer now executes the halted step through the ledger entry's own resume handle; the documented ratify-then-resume flow no longer dies on the single-use ledger, and no interleaving can double-run a material step (260819-c55)
|
|
50
|
+
- Part 8 hook matcher extended to the pws-brain-mcp server naming - the egress guard and response sanitizer no longer no-op on live Brain calls
|
|
51
|
+
- Phase 234-01 instructions pin evolved from absence to byte-identity: the initialize channel serves exactly RUNTIME_INSTRUCTIONS, moat-vocabulary canaries asserted absent
|
|
52
|
+
|
|
53
|
+
### Changed
|
|
54
|
+
- Every live surface (setup templates, brain-setup docs, THE-BRAIN, WORKFLOWS, CANON glossary, session-start banner, .env.brain.template) points at the canon store pws-brain-mcp.onrender.com; Aura-era references marked RETIRED do-not-build-against; the stale replica is suspended Brain-side and answers 503
|
|
55
|
+
|
|
56
|
+
## [2.0.0-beta.5] - 2026-08-11
|
|
57
|
+
|
|
58
|
+
### Added
|
|
59
|
+
-
|
|
60
|
+
|
|
61
|
+
## [2.0.0-beta.3] - 2026-08-11
|
|
62
|
+
|
|
63
|
+
### Added
|
|
64
|
+
-
|
|
65
|
+
|
|
66
|
+
### Fixed
|
|
67
|
+
- **The doctor's statusline self-test no longer cries wolf on Windows.** `/mos:doctor` used to
|
|
68
|
+
spawn the plugin's `scripts/statusline-mos` script directly, skipping the `bash` wrapper that
|
|
69
|
+
the real session always uses (`settings.json`'s `statusLine.command` is
|
|
70
|
+
`bash "${CLAUDE_PLUGIN_ROOT}/scripts/statusline-mos"`). Windows has no direct-exec association
|
|
71
|
+
for an extensionless bash-shebang script, so the self-test reported a "synthetic spawn error"
|
|
72
|
+
even on a machine where the statusline itself was rendering correctly -- caught live on a group
|
|
73
|
+
Windows install call, where a tester's own diagnosis nailed it: "It's calling the script
|
|
74
|
+
directly instead of through Bash which doesn't work on Windows regardless of whether the status
|
|
75
|
+
line itself is healthy. That's a diagnostic bug, not a problem." The self-test now routes
|
|
76
|
+
through `bash` whenever the effective statusline command is the plugin's own bash script (or a
|
|
77
|
+
user override that itself starts with `bash`), matching the real runtime exactly.
|
|
78
|
+
- **`/mos:update` no longer gives up on the first network hiccup.** The version checker fetched
|
|
79
|
+
both its lookup URLs from a single host (`raw.githubusercontent.com`) with no retry, so one
|
|
80
|
+
transient reset failed the whole check and printed `STATUS=NETWORK_ERROR` -- reported from a
|
|
81
|
+
field session on an institutional network (karunya.edu.in) where `/mos:update` looped on
|
|
82
|
+
`read ECONNRESET` every time. It now retries up to three times with 250/500/1000ms backoff, but
|
|
83
|
+
only for genuinely transient errors (`ECONNRESET`, `ETIMEDOUT`, `EAI_AGAIN`); a real HTTP error
|
|
84
|
+
still fails immediately so a broken fetch doesn't stall behind a pointless retry.
|
|
85
|
+
Follow-up not yet built: auto-degrading straight to the native `/plugin update` path once
|
|
86
|
+
retries exhaust, instead of handing the user two commands to paste.
|
|
87
|
+
|
|
1
88
|
## [2.0.0-beta.1] - 2026-08-10
|
|
2
89
|
|
|
3
90
|
### Fixed
|
package/README.md
CHANGED
|
@@ -7,10 +7,10 @@
|
|
|
7
7
|
|
|
8
8
|
A thinking partner that sits above your AI. You talk through a problem worth solving. Larry finds the real problem before you solve the wrong one, brings the right method at the moment you need it, pushes back when your confidence outruns your evidence, and turns the conversation into a structured room that remembers every decision and catches what you missed.
|
|
9
9
|
|
|
10
|
-
Powered by PWS (Problems Worth Solving), an innovation methodology built and tested through 20 years of teaching
|
|
10
|
+
Powered by PWS (Problems Worth Solving), an innovation methodology built and tested through 20 years of teaching.
|
|
11
11
|
Engineered by Jonathan Sagir.
|
|
12
12
|
|
|
13
|
-
[](CHANGELOG.md)
|
|
14
14
|
[](LICENSE)
|
|
15
15
|
[](#three-surfaces)
|
|
16
16
|
|
|
@@ -21,82 +21,49 @@
|
|
|
21
21
|
|
|
22
22
|
---
|
|
23
23
|
|
|
24
|
-
##
|
|
24
|
+
## You have a problem worth solving. You are probably solving the wrong one first.
|
|
25
25
|
|
|
26
|
-
|
|
27
|
-
|
|
28
|
-
You never learn a tool. You talk. The room takes shape underneath the conversation. Commands exist, but they are internals, and Larry routes you to them when they help. Whatever you said yesterday is still working for you today.
|
|
29
|
-
|
|
30
|
-
---
|
|
31
|
-
|
|
32
|
-
## Four jobs it does for you
|
|
33
|
-
|
|
34
|
-
- **Find the problem.** Larry reframes what you are working on before you try to solve it, so you spend your effort on the real problem, not the first one you saw.
|
|
35
|
-
- **Show what you missed.** Every new entry is compared against everything already in the room. Contradictions, gaps, and cross-project connections surface on their own.
|
|
36
|
-
- **Build what you could not alone.** The right framework at the right stage, chained intelligently, across more domains than one mind can hold at once.
|
|
37
|
-
- **Defend what you decided.** Every decision and the reason behind it becomes working memory you can stand behind weeks later.
|
|
38
|
-
|
|
39
|
-
---
|
|
40
|
-
|
|
41
|
-
## How it works (three pieces)
|
|
42
|
-
|
|
43
|
-
### Larry is the product
|
|
44
|
-
|
|
45
|
-
Larry is the AI you talk to, and the conversation is the whole surface. Larry asks the questions that reframe the problem before you try to solve it, brings the method that fits where you are, and files what you say into your room without making you stop to organize. You do not have to know any framework names. You describe what you are doing. Larry routes you. There is nothing else to learn.
|
|
46
|
-
|
|
47
|
-
### The room is the receipt
|
|
48
|
-
|
|
49
|
-
You talk; the room writes itself. Every conversation, every meeting, every decision lands in a folder structure organized by venture stage: the problem, the market, the solution, the team, the money, the IP, the meetings, the opportunities. You open it in your file manager. You back it up like any other folder. You own it. You never filed a thing.
|
|
50
|
-
|
|
51
|
-
### The room surfaces what you cannot see
|
|
52
|
-
|
|
53
|
-
Every time you add something new, the system compares it against everything already there. Larry tells you what just changed, what contradicts what, what connects to what, and what is now missing. You decide: APPROVE, REJECT (with a reason), or DEFER. The reason becomes part of the room. The next scan is smarter. When a scan surfaces an opportunity, it does not stop at a headline: you qualify it at a card, and one explicit Explore turns it into cited deep research filed in your opportunity bank. The web works the same way: paste a URL and, once you approve, the page is filed as a cited source in your room and compared against everything already there. When something breaks mid-research, it tells you exactly what happened and what it tried next, never a silent empty result.
|
|
26
|
+
That is the job MindrianOS is hired for. Not "take my notes" and not "chat with an AI." You bring a real problem, a venture, a research question, a decision you cannot get right alone, and Larry's first move is almost always a question that reframes it, because the version of the problem you walked in with is rarely the one worth solving. Finding the right method, catching what you missed, and remembering what you decided all follow from getting that first reframe right.
|
|
54
27
|
|
|
55
28
|
---
|
|
56
29
|
|
|
57
|
-
##
|
|
58
|
-
|
|
59
|
-
You can. MindrianOS is not instead of them. It is the layer that makes one of them remember, judge, and hold you to your own reasoning.
|
|
60
|
-
|
|
61
|
-
A raw AI is brilliant for one turn and forgets the next. Ask it the same venture question next week and it relearns you from scratch. It knows every framework but not which one you need right now. And it is agreeable: it will help you build a beautiful deck on a false premise, because it optimizes for a helpful answer, not a true one.
|
|
30
|
+
## The loop, in 30 seconds
|
|
62
31
|
|
|
63
|
-
|
|
32
|
+
This is the whole mental model. You do not need more than this to use MindrianOS well.
|
|
64
33
|
|
|
65
|
-
|
|
66
|
-
|
|
67
|
-
|
|
68
|
-
|
|
34
|
+
1. **You talk.** Whatever is in your room right now, the venture, the decision, the meeting you just filed, becomes context.
|
|
35
|
+
2. **Your context triggers a question to the Brain**, the methodology graph: 28,325 nodes and 181 frameworks built from 20 years of teaching, holding WHEN to use WHICH method and in WHAT sequence.
|
|
36
|
+
3. **Larry joins the answer to your situation.** Not a lecture pulled from a textbook. Your problem, run through real methodology.
|
|
37
|
+
4. **You ratify what matters.** Approve it, reject it with a reason, or defer it. Your call becomes part of the room.
|
|
38
|
+
5. **The room remembers.** Next time, it is already there, checking today's input against it.
|
|
69
39
|
|
|
70
|
-
|
|
40
|
+
When the graph genuinely has nothing structured for what you asked, Larry says so plainly instead of making something up, and queues the gap for enrichment. He never improvises methodology. See "What an honest refusal looks like" below: that is not an error message, it is the whole point.
|
|
71
41
|
|
|
72
42
|
---
|
|
73
43
|
|
|
74
44
|
## Install
|
|
75
45
|
|
|
76
|
-
|
|
77
|
-
|
|
78
|
-
### npm (one line, recommended)
|
|
46
|
+
Three commands. Full walkthrough at [the install guide](https://mindrian-os.com/docs/install).
|
|
79
47
|
|
|
80
48
|
```bash
|
|
81
49
|
npx @mindrian_os/cli
|
|
82
50
|
```
|
|
83
51
|
|
|
84
|
-
|
|
52
|
+
Or, from inside Claude Code:
|
|
85
53
|
|
|
86
54
|
```bash
|
|
87
55
|
claude plugin marketplace add jsagir/mindrian-marketplace
|
|
88
56
|
claude plugin install mos@mindrian-marketplace
|
|
89
57
|
```
|
|
90
58
|
|
|
91
|
-
|
|
59
|
+
Restart Claude Code. Larry starts talking, and your install quietly registers its own Brain identity in the background: no API key to paste, no account to create first. If you already have a Brain key, it wins and nothing changes.
|
|
92
60
|
|
|
93
|
-
|
|
94
|
-
curl -sL https://raw.githubusercontent.com/jsagir/mindrian-os-plugin/main/install.sh | bash
|
|
95
|
-
```
|
|
61
|
+
Two things the field taught us, worth checking before you start:
|
|
96
62
|
|
|
97
|
-
|
|
63
|
+
- You need Claude Pro or Max on your own personal account. A company-managed (SSO or Okta) Claude plan blocks the in-app upgrade prompt; a personal account does not.
|
|
64
|
+
- On Windows, the Node.js installer offers an optional "Tools for Native Modules" checkbox. Leave it unchecked. MindrianOS ships no native modules, and checking it triggers a long, unrelated Visual Studio Build Tools install.
|
|
98
65
|
|
|
99
|
-
|
|
66
|
+
Update or repair an install:
|
|
100
67
|
|
|
101
68
|
```bash
|
|
102
69
|
mindrian-os update # marketplace + plugin update
|
|
@@ -107,40 +74,39 @@ A note on install prompts: Claude Code asks you to approve each shell command. 1
|
|
|
107
74
|
|
|
108
75
|
---
|
|
109
76
|
|
|
110
|
-
## What
|
|
77
|
+
## What talking to Larry feels like
|
|
111
78
|
|
|
112
|
-
|
|
79
|
+
Most of the time it feels like a sharp colleague who happens to know 181 frameworks and has read your whole project. You ask, Larry answers through the loop above, and a graph-grounded answer carries a source line so you know where it came from:
|
|
113
80
|
|
|
114
|
-
|
|
81
|
+
> ■ BRAIN: Jobs to Be Done · framework · readiness 4/4
|
|
115
82
|
|
|
116
|
-
|
|
117
|
-
|
|
118
|
-
|
|
119
|
-
|
|
120
|
-
/
|
|
121
|
-
/mos:bono # a six-hats research-and-debate team on your question
|
|
122
|
-
/mos:map-unknowns # hunt the claims you are most confident about, and wrong
|
|
123
|
-
/mos:file-meeting # paste a transcript, Larry files it
|
|
124
|
-
/mos:research <url> # paste a link, approve the card, the page becomes cited room knowledge
|
|
125
|
-
/mos:opportunities # what grants match this room right now
|
|
126
|
-
/mos:qualify-opportunity # judge a surfaced opportunity at a card; Explore turns it into research
|
|
127
|
-
/mos:graph "what is the weakest assumption in my financial model?"
|
|
128
|
-
/mos:grade # honest assessment against real ventures
|
|
129
|
-
```
|
|
83
|
+
A conversation turn, Larry thinking out loud with you rather than consulting the graph, carries no source line. The absence is the signal: no line means it is talk, not method.
|
|
84
|
+
|
|
85
|
+
Sometimes the graph does not have what you need yet, and that is not hidden from you. Here is a real refusal, verbatim:
|
|
86
|
+
|
|
87
|
+
> The graph doesn't have Jobs to Be Done structured yet (readiness 2/4; missing: examples, edge-cases). I've queued it for enrichment. I can share what the graph does hold on this, marked as partial, or we work without it.
|
|
130
88
|
|
|
131
|
-
That is a
|
|
89
|
+
That is a feature, not an outage. A tool that quietly guesses when it does not know is worse than one that tells you and keeps a list. A keyless or unreachable session gets the same treatment: an honest refusal and a visible path forward, never an imitation of an answer it does not have.
|
|
132
90
|
|
|
133
91
|
---
|
|
134
92
|
|
|
135
|
-
##
|
|
93
|
+
## The three layers
|
|
136
94
|
|
|
137
|
-
|
|
95
|
+
| Layer | What | Who owns it |
|
|
96
|
+
|-------|------|-------------|
|
|
97
|
+
| **Plugin** | Skills, commands, agents, and hooks that run the conversation | Open, in this repo |
|
|
98
|
+
| **Brain** | The methodology graph: 28,325 nodes, 181 frameworks, 20 years of teaching, served over MCP | Served remotely, never distributed |
|
|
99
|
+
| **Room** | Your venture, your decisions, your files | Yours, on your machine, always |
|
|
138
100
|
|
|
139
|
-
|
|
101
|
+
The Brain never sees your room. Every query it answers carries a generic methodology question, never your notes, your decisions, or your meetings.
|
|
140
102
|
|
|
141
|
-
|
|
103
|
+
---
|
|
104
|
+
|
|
105
|
+
## Why not just talk to Claude, ChatGPT, or Gemini directly?
|
|
142
106
|
|
|
143
|
-
|
|
107
|
+
You can. MindrianOS is not instead of them, it is the layer that makes one of them remember, judge, and hold you to your own reasoning. A raw AI is brilliant for one turn and forgets the next. It knows every framework but not which one you need right now. And it is agreeable: it will help you build a beautiful deck on a false premise, because it optimizes for a helpful answer, not a true one.
|
|
108
|
+
|
|
109
|
+
MindrianOS adds what a raw model structurally cannot be: it remembers (your room persists across every session), it knows when (decades of teaching calibrate which method fits which stage), it pushes back (a co-founder tells you when you are wrong, a chatbot tells you that you are brilliant), and it keeps your data yours (only generic methodology crosses to the Brain, never your specifics).
|
|
144
110
|
|
|
145
111
|
---
|
|
146
112
|
|
|
@@ -156,46 +122,43 @@ MindrianOS works wherever Claude works. Same Larry, same room, every surface.
|
|
|
156
122
|
|
|
157
123
|
---
|
|
158
124
|
|
|
159
|
-
##
|
|
125
|
+
## Commands are internals
|
|
160
126
|
|
|
161
|
-
|
|
127
|
+
Talk. That is the whole interface. Larry reaches for the right command on your behalf; the ones below are here for when you already know the shortcut.
|
|
162
128
|
|
|
163
|
-
|
|
129
|
+
```bash
|
|
130
|
+
/mos:ignite # the front door: start or excavate a room
|
|
131
|
+
/mos:discover # Larry-led client + product + JTBD discovery
|
|
132
|
+
/mos:beautiful-question # reframe the problem before solving it
|
|
133
|
+
/mos:file-meeting # paste a transcript, Larry files it
|
|
134
|
+
/mos:graph "what is the weakest assumption in my financial model?"
|
|
135
|
+
/mos:grade # honest assessment against real ventures
|
|
136
|
+
```
|
|
164
137
|
|
|
165
|
-
|
|
138
|
+
That is a slice of over a hundred commands across the skills, agents, and pipelines this plugin ships. If you do not know which one to run, that is the normal case. Just talk.
|
|
166
139
|
|
|
167
140
|
---
|
|
168
141
|
|
|
169
142
|
## Pricing
|
|
170
143
|
|
|
171
|
-
Free plugin. It requires a paid Claude plan (Claude Pro, $20/mo, or higher) because it runs on top of Claude. The Brain
|
|
144
|
+
Free plugin. It requires a paid Claude plan (Claude Pro, $20/mo, or higher) because it runs on top of Claude. The Brain installs with it, silently registered, at no separate cost.
|
|
172
145
|
|
|
173
146
|
---
|
|
174
147
|
|
|
175
148
|
## The privacy line
|
|
176
149
|
|
|
177
|
-
MindrianOS reads your workspace and writes only to your rooms (default: `~/MindrianRooms/`) and to session state (`./.mindrian/`). It does not push anything to the Brain
|
|
150
|
+
MindrianOS reads your workspace and writes only to your rooms (default: `~/MindrianRooms/`) and to session state (`./.mindrian/`). It does not push anything to the Brain beyond a generic methodology question: never your notes, never your decisions, never your meetings.
|
|
178
151
|
|
|
179
152
|
For zero permission prompts during a session: `claude --dangerously-skip-permissions`. The read/write surface is bounded to your workspace and your rooms. For granular control, copy the matcher set from [`docs/settings-template.json`](docs/settings-template.json) into `~/.claude/settings.json`.
|
|
180
153
|
|
|
181
154
|
---
|
|
182
155
|
|
|
183
|
-
## Why PWS, why Larry
|
|
184
|
-
|
|
185
|
-
PWS (Problems Worth Solving) is not a checklist. It is a way of thinking about ventures as wicked problems that need to be reframed before they can be solved, and that demand a working memory because nobody can hold the whole thing in their head.
|
|
186
|
-
|
|
187
|
-
Larry is the personality that delivers PWS in your terminal, and the teaching is intrinsic. You do not have to know the framework names. Larry asks the question, suggests the move, shows the chain. You decide.
|
|
188
|
-
|
|
189
|
-
---
|
|
190
|
-
|
|
191
156
|
## Links
|
|
192
157
|
|
|
193
158
|
- **Website**: [mindrian-os.com](https://mindrian-os.com)
|
|
194
159
|
- **Marketplace**: [github.com/jsagir/mindrian-marketplace](https://github.com/jsagir/mindrian-marketplace)
|
|
195
160
|
- **Changelog**: [CHANGELOG.md](CHANGELOG.md)
|
|
196
|
-
- **Brain
|
|
197
|
-
- **PWS, Prof. Lawrence Aronhime**: [LinkedIn](https://www.linkedin.com/in/lawrence-aronhime-8363894/)
|
|
198
|
-
- **Jonathan Sagir**: [LinkedIn](https://www.linkedin.com/in/jonathansagir/)
|
|
161
|
+
- **Brain key (override path)**: [Request one](https://mindrian-os.com/brain-access)
|
|
199
162
|
|
|
200
163
|
---
|
|
201
164
|
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@mindrian_os/cli",
|
|
3
|
-
"version": "2.0.0-beta.
|
|
3
|
+
"version": "2.0.0-beta.13",
|
|
4
4
|
"description": "Install MindrianOS into Claude Code with one command -- `npx @mindrian_os/cli`. Ships the MindrianOS plugin (Larry + PWS methodology + Data Room) plus a setup/diagnostics CLI (install/doctor/update).",
|
|
5
5
|
"scripts": {
|
|
6
6
|
"mcp": "node bin/mindrian-mcp-server.cjs",
|