This console: crawls GitHub trending daily, Claude-reads each new repo, and routes what applies to which project.
Progression
Timeline
Tree
start roadmap →
NextApply KnockOutEZ/wigolo: Local-first web intelligence engine for AI agents — search, fetch, crawl, and research without planning plan
2026-07-16shippedProject file browser panel (VS Code-style tree, open + copy path)
2026-07-18shippedProject feature tree (skill-tree progression view)
2026-08-10shippedAI Companion — confirm-gated dispatcher + cross-project reporting
2026-07-21shippedDepth-bounded recursive research mode (query → deepen → synthesize)
2026-07-21shippedTopic-centric multi-source synthesis with per-claim provenance
2026-07-21shippedCompanion: debounce/buffer voice input before acting
2026-07-21shippedCompanion: voice-confirm (approve in-channel) as the dispatch wait-sta…
2026-07-21shippedPipeline hardening: bounded retry/backoff + AI-output sanity gate
2026-08-02shippedHost Claude Code sessions inside the console — one surface, no per-pro…
2026-08-03shippedAdopt ceo-sim's parallel-agent machinery: rebase-before-merge, file-co…
2026-08-06shippedLet a session choose its reasoning effort, and change it on a live one
2026-08-06shippedRun a plan as a relay: one session, a different role per task
2026-08-07shippedPresent the result automatically, and open it after you say yes
2026-08-10shippedThe console reads as the path: chrome in the companion register, Ledge…
2026-08-17shippedOne profile record, many destinations. A PUBLISH-step surface holding …
Needs review
Lessons
approach
proposed
Start the console in the same shell call as the suite
approach
proposed
Verify MCP server presence/absence from the SDK init message's mcp_servers array (name+status), never the tools array.
approach
proposed
No data-plugin OAuth MCP server is authorized on this machine, so re-declaration-keeps-auth cannot be tested without an interactive OAuth flow first.
stop reason
proposed
The run stopped: task 1 did not report a state — nothing to advance to
approach
proposed
Before running a researcher probe task, check git log for the plan's named output file — it may already be done and merged.
stop reason
proposed
The run stopped: a task reported failed
task failed
proposed
Task 1 failed: Probe strict_mcp_config against a live session — does it strip PLUGIN servers?
approach
proposed
Task 1's exact deliverable (docs/mcp-strict-scoping-probe.md) was already produced and committed by a prior session before this dispatch started
constitution friction
proposed
services/mcp.py (built earlier by commit 5a0bf23, for an unrelated Setup/Config-view feature) already exists and already looks like task 5's deliverable, but its plugin enumeration silently drops 2 of the 3 currently-enabled plugin MCP sources
constitution friction
proposed
Fixing the external_plugins/ nesting alone does not get context7+playwright both visible — playwright's .mcp.json has no top-level mcpServers key, the file itself IS the server map ({"playwright": {...}}), a second distinct bug in plugin .mcp.json parsing
empty done
proposed
Task 1 was recorded done unverified: Probe strict_mcp_config against a live session — does it strip PLUGIN servers?
task failed
proposed
Task 3 failed: Contract: the mcp key on the constitution, the resolver seam, the project route
empty done
proposed
Task 3 was recorded done unverified: Contract: the mcp key on the constitution, the resolver seam, the project route
task failed
proposed
Task 4 failed: tests/test_mcp.py: the resolver, red first
approach
proposed
validate() checking mcp names against mcp.available() must not run inside tests that build resolve()-fixture constitutions via constitution.save() — available() has no home/projects override and always resolves against the real machine
empty done
proposed
Task 4 was recorded done unverified: tests/test_mcp.py: the resolver, red first
empty done
proposed
Task 5 was recorded done unverified: Add the resolver to services/mcp.py
task failed
proposed
Task 7 failed: Extend tests/test_session.py for the options contract
approach
proposed
Session.as_row() names the resolved-server-count field mcp_count, not mcp — docs/mcp-scoping-contract.md#4 says the row gains "mcp": <int|None>, but task 7's own instructions said mcp_count, and task 7's instructions win since they're the literal ask handed to the implementer.
approach
proposed
The contract doc says tasks 4/5 added a 'key' field to services/mcp.py::servers() rows. They did not — and task 9 cannot build the picker without it, since every checkbox value and the auth-cache join are keyed on it.
stop reason
proposed
The run stopped: task 9 did not report a state — nothing to advance to
approach
proposed
In a worktree session, the documented browser check silently validates the OTHER checkout — pass DESOUL_UI_BASE at a console started from the worktree, or the run proves nothing.
stop reason
proposed
The run stopped: task 14 did not report a state — nothing to advance to
empty done
proposed
Task 9 was recorded done unverified: Wire the MCP block: template, POST route, and the project JSON route
Nothing here changed a prompt until you kept it. A kept lesson retires after 90 days unless a later run re-earns it with fresh evidence.
Timeline
2026-07-16done
Plan “Project file browser panel (VS Code-style tree, op…” — Shipped file browser: new services/filetree.py (stdlib, one-level lazy walk, .gitignore+skip-set, path-escape guarded, 40KB text/binary preview). app.py +3 endpoints (/tree, /file, /open code shell-out) +1 import. project_detail.html Files panel (lazy tree, type-filter, copy-abs-path, open-in-editor, inline <pre> preview). style.css .ft-* warm-paper styles. Tests: 7/7 end-to-end pass (lazy load, gitignore/skip, path-escape 400, text+binary preview, no-hang) + open-endpoint guards. Constitution note: app.py is 'ask before editing' — edited because task 2 requires endpoints and all routes live there; kept to 3 endpoints + 1 import (4 files total, under max 8).
2026-07-18done
Plan “Project feature tree (skill-tree progression view)” — Shipped feature tree (skill-tree progression): new services/feature_tree.py + pipeline/tree_build.py (git-gated daily) + prompts/extract_tree.md; Timeline/Tree toggle + SVG renderer (console.js) + Ledger styles. 9 files (4 new / 5 edited incl. app.py). Tests 195 pass (12 new). Verified end-to-end: real build of this repo → 18 nodes/5 groups, done-verified vs repo paths, renders at ?view=tree. Constitution exception: app.py (ask-before-edit) edited minimally — 1 import + view param + 2 endpoints (routes must live there, per file-browser precedent).
2026-08-10done
Plan “AI Companion — confirm-gated dispatcher + cross-pr…” — 1-9 shipped between July and 2026-08-05; their states were never recorded, so an approved plan sat on the tracker looking untouched — the exact defect the landing gate (five-gaps task 3) exists to prevent. Ticked from evidence and named as such. 10 and 11 move into the follow-on plan.
2026-07-21done
Plan “Depth-bounded recursive research mode (query → dee…” — Delivered together as services/deepdive.py + /deepdive: bounded recursion (#1) + per-claim provenance & refuse-on-missing (#2). Local corpus, subscription runner.
2026-07-21done
Plan “Topic-centric multi-source synthesis with per-clai…” — Delivered together as services/deepdive.py + /deepdive: bounded recursion (#1) + per-claim provenance & refuse-on-missing (#2). Local corpus, subscription runner.
2026-07-21done
Plan “Companion: debounce/buffer voice input before acti…” — Client-side buffer+debounce in the companion controller: transcriptions accumulate, finalize after ~2.5s pause or Prepare.
2026-07-21done
Plan “Companion: voice-confirm (approve in-channel) as t…” — Voice-confirm: TTS read-back then strict-affirmative listen (default-deny) triggers the existing confirm=true launch; click remains fallback.
2026-07-21done
Plan “Pipeline hardening: bounded retry/backoff + AI-out…” — retry.py (backoff, injectable sleep) wraps README fetch + claude_runner timeout; categorize.sanitize_entry coerces bad verdict/apply_where before write. Tests: test_retry + test_categorize.
2026-08-02done
Plan “Host Claude Code sessions inside the console — one…” — Shipped 2026-08-02 across 12e2aed..fccecf8. Sessions run in-process on the subscription via claude-agent-sdk, gated by a PreToolUse hook that holds even at bypassPermissions. Held calls are answerable (allow / always / deny), with always-allow writing to the PROJECT constitution. Writing sessions get a worktree; you read the diff and merge on a button, and a conflict refuses rather than resolving. One needs-you count in the global chrome. Terminal.app launch removed, no escape hatch; docs/agentic-os-decision.md records the reversal and its reason. Found along the way: system_prompt defaulted to empty, so sessions never knew their own working directory — that one bug explained the stray home-directory reads, the pwd probes, and much of the ask noise.
2026-08-03done
Plan “Adopt ceo-sim's parallel-agent machinery: rebase-b…” — All 5 tasks shipped 2026-08-03. Rebase-before-merge at every turn end with an 'ask it to resolve' handoff; target-file prediction + collision scheduling replacing the flat refusal; manual trust tiers (intern/mid/senior/principal) that silence classes of ask and can never reach a deny. Auto-advancement deliberately dropped — no honest ratchet signal exists without the console running each project's tests. Found live: silencing the guard alone is not enough, because the model reads the constitution and stops in prose before the guard ever sees the call; the rules block itself now reflects the tier.
2026-08-06done
Plan “Let a session choose its reasoning effort, and cha…” — All 5 tasks done. session.py: EFFORTS + session_effort() + converse(effort=) with NO fallback (unlike model) so plans/dispatch/deepdive stay unchanged. session_registry.py: Session.effort, start(effort=), and change_effort() (close+wait+resume — no set_effort exists). app.py: Form param on /api/sessions, sessions_view context, new POST /api/sessions/{id}/effort (404/400/409). sessions.html: Effort select, opts-summary, header meta, 'Restart at effort' control with a confirm() explaining it restarts the session. 7 files touched (within the 8-file cap), TDD throughout. Full suite 671 passed / 1 pre-existing unrelated failure (doc-rot referencing a gitignored data file). Browser-verified on a temporary instance (port 8799, torn down after) since the live :8790 console is a separate long-running process outside this worktree. One documented exception: app.py edits were blocked for one turn by a PreToolUse hook timeout unrelated to the constitution itself (resolved on retry, noted to the user directly). Independent reviewer confirmed no invented SDK surface and no silent default change.
2026-08-06done
Plan “Run a plan as a relay: one session, a different ro…” — Relay shipped: a plan runs as one session, a role per task. 8 new tests, full suite 693 green.
2026-08-07done
Plan “Present the result automatically, and open it afte…” — The result presents itself and the yes acts: merge opens what was made, from a derived allowlist. 33 new tests, suite 726 green.
2026-08-10done
Plan “The console reads as the path: chrome in the compa…” — 7/7. The console reads as the path: the stranger test's three failures are each fixed with browser-measured evidence, /reports is the one recorded exception and it is already scheduled with an answered clarification. The human half — a real stranger watching the film — stays with Le.
2026-08-17done
Plan “One profile record, many destinations. A PUBLISH-s…” — Built the Profile surface. state/profile.yaml is now the source of truth for the CV, the portfolio, the Upwork overview and the LinkedIn About; services/profile.py loads, refuses an inconsistent record, re-derives measurable numbers and renders each destination by region replacement so hand-designed markup is never regenerated. /profile lives under PUBLISH and joined docs/spine.md in the same task. career-ops installed for the inbound side, per the third clarification.
SIX DRIFT BUGS FOUND, none of them by eye: 1,164->1,217 repositories, 112->116 test modules, ~30k->~32k lines, a facts grid my own renderer silently cut 5->3, two markers wrapped outside their container so the first render ate <p class="sub">, and the Upwork hook restating a number as typed text rather than referencing it.
Task 6's verdict, measured end to end on the real files: one edit moved 4/4 surfaces with no second edit anywhere. It failed first, on Upwork, which is what made it worth running.
Full suite 1464 passed. Constitution grants exactly two paths outside this repo.
Derived from applied history + mined skills + completed plans. Ground truth stays in the project's git log; this is the index.
Outcome
● up · 1137ms
data/pipeline.log · (log file not found)
data/console.log · (log file not found)
Content room
○ not running
Workflows
Run full pipeline
Run test suite
idle
Brainstorm
open space →
Think a feature or fix through with the advisor → promote to a plan.
Constitution
edit rules & trust tier →
trust tier
senior · custom rules · injected into every apply & execution promptMatch rules
Applied history · 63 — restore / mine a skill
Dismissed · 198 — reviewed, no-take (restore to re-queue)
·
sourcebot-dev/sourcebot◑ WATCH
·
·
smart-mcp-proxy/mcpproxy-go◑ WATCH
·
steipete/CodexBar◑ WATCH
·
statewright/statewright◑ WATCH
·
superdoc-dev/superdoc○ IGNORE
·
tadas-github/a2asearch-mcp◑ WATCH
·
sxhxliang/mcp-access-point◑ WATCH
·
tap2k/modelun◑ WATCH
·
tigranbs/mcgravity◑ WATCH
·
thinkchainai/mcpbundles○ IGNORE
·
theopenco/llmgateway◑ WATCH
·
tirth8205/code-review-graph◆ SUBSCRIBE
·
traceloop/openllmetry◑ WATCH
·
tsouth89/toolport◑ WATCH
·
vaquarkhan/mcp-test-harness◑ WATCH
·
upstash/context7◆ SUBSCRIBE
·
vinkius-labs/vurb.ts◑ WATCH
·
wbopan/retro-harness◑ WATCH
·
whiteknightonhorse/APIbase○ IGNORE
·
wolfoo2931/linkedrecords○ IGNORE
·
·
wandb/weave○ IGNORE
·
sitbon/magg◑ WATCH
·
shareAI-lab/learn-claude-code◆ SUBSCRIBE
·
sgl-project/sglang◑ WATCH
·
sentrux/sentrux◑ WATCH
·
scotia1973-bot/api-hub◑ WATCH
·
sF1nX/x402station-mcp◑ WATCH
·
rupinder2/mcp-orchestrator◑ WATCH
·
rplryan/x402-discovery-mcp◑ WATCH
·
ronak-create/FableCut◑ WATCH
·
·
robhunter/agentdeals◑ WATCH
·
revfactory/harness◑ WATCH
·
reefapi/reefapi-mcp◑ WATCH
·
raphaelchristi/harness-evolver◆ SUBSCRIBE
·
quokkapix/quokkapix-mcp◑ WATCH
·
pydantic/logfire◑ WATCH
·
puppeteer/puppeteer◑ WATCH
·
protocolbuffers/protobuf◑ WATCH
·
profullstack/mcp-server◑ WATCH
·
portel-dev/ncp◑ WATCH
·
pollinations/pollinations◑ WATCH
·
particlefuture/1mcpserver◑ WATCH
·
onesuper/tui-use◑ WATCH
·
mvanhorn/last30days-skill◑ WATCH
·
·
·
modelcontextprotocol/inspector◆ SUBSCRIBE
·
mksglu/context-mode◆ SUBSCRIBE
·
mindfold-ai/Trellis◑ WATCH
·
microsoft/skills-for-fabric◑ WATCH
·
microsoft/playwright-mcp◑ WATCH
·
microsoft/LLMLingua◑ WATCH
·
metatool-ai/metatool-app◑ WATCH
·
merterbak/Grok-MCP◑ WATCH
·
mem0ai/mem0◑ WATCH
·
matt1398/claude-devtools◆ SUBSCRIBE
·
mambalabsdev/mcp-gtm-suite◑ WATCH
·
lyogavin/airllm◑ WATCH
·
lightseekorg/tokenspeed◑ WATCH
·
langchain-ai/openwiki◑ WATCH
·
kvcache-ai/ktransformers◑ WATCH
·
khalidsaidi/ragmap◑ WATCH
·
·
juspay/neurolink◑ WATCH
·
julien040/anyquery◑ WATCH
·
jiji262/douyin-downloader◑ WATCH
·
jbwinters/jacquard-lang◑ WATCH
·
·
·
·
jamiepine/voicebox◑ WATCH
·
isaac-levine/forage◑ WATCH
·
ikoskela/wisepanel-mcp◑ WATCH
·
ibelick/ui-skills◑ WATCH
·
iOfficeAI/OfficeCLI◑ WATCH
·
hoangnb24/harness-experimental◆ SUBSCRIBE
·
·
hefengcs/HIVE◑ WATCH
·
·
·
·
gzoonet/cortex◑ WATCH
·
gregario/astronomy-oracle◑ WATCH
·
gpu-bridge/mcp-server◑ WATCH
·
google-labs-code/design.md◑ WATCH
·
glenngillen/mcpmcp-server◑ WATCH
·
github/github-mcp-server◑ WATCH
·
garrytan/gstack◑ WATCH
·
future-agi/future-agi◑ WATCH
·
forgemeshlabs/anomaly-mcp◑ WATCH
·
firecrawl/open-lovable◑ WATCH
·
fastapi/fastapi◑ WATCH
·
ertad-family/liquid◆ SUBSCRIBE
·
entire-vc/evc-spark-mcp◑ WATCH
·
elder-plinius/G0DM0D3◑ WATCH
·
edgarriba/prolink◑ WATCH
·
dottxt-ai/outlines◑ WATCH
·
depwire/depwire◑ WATCH
·
decolua/9router◆ SUBSCRIBE
·
danielmiessler/LifeOS◑ WATCH
·
crynta/terax-ai◑ WATCH
·
coollabsio/coolify◑ WATCH
·
comet-ml/opik◑ WATCH
·
codeislaw101/katzilla◑ WATCH
·
chopratejas/headroom◆ SUBSCRIBE
·
·
cfpramod/open-museum-mcp◑ WATCH
·
·
·
callstack/agent-device◑ WATCH
·
burningion/video-editing-mcp◑ WATCH
·
bradautomates/claude-video◑ WATCH
·
blockrunai/blockrun-mcp◑ WATCH
·
backnotprop/plannotator◑ WATCH
·
awslabs/agentcore-samples◑ WATCH
·
austenstone/myinstants-mcp◑ WATCH
·
attalla1/photopea-mcp-server◑ WATCH
·
asmith26/jupytercad-mcp◑ WATCH
·
arikusi/nakkas◑ WATCH
·
arikusi/deepseek-mcp-server◑ WATCH
·
·
apify/crawlee◑ WATCH
·
anthropics/skills◑ WATCH
·
·
anthropics/claude-cookbooks◆ SUBSCRIBE
·
alexzavialov/travel-art-mcp◑ WATCH
·
alexar76/aimarket-plugins◑ WATCH
·
·
albertnahas/icogenie-mcp◑ WATCH
·
·
ahujasid/blender-mcp◑ WATCH
·
agentbodegastore/agentbodega◑ WATCH
·
affaan-m/everything-claude-code◆ SUBSCRIBE
·
adw0rd/awesome-mcp-tools-mcp◑ WATCH
·
aattaran/deepclaude◑ WATCH
·
·
Work90210/APIFold◆ SUBSCRIBE
·
ViperJuice/mcp-gateway◑ WATCH
·
·
Syncause/debug-skill◑ WATCH
·
·
SuperagenticAI/metaharness◑ WATCH
·
SigNoz/signoz◑ WATCH
·
Rendeverance/toolfunnel◑ WATCH
·
ROCm/TheRock◑ WATCH
·
PrismML-Eng/Bonsai-demo◑ WATCH
·
PostHog/posthog◑ WATCH
·
Paca-AI/paca◑ WATCH
·
OthmanAdi/planning-with-files◆ SUBSCRIBE
·
OpenSenseNova/SenseNova-U1◑ WATCH
·
Octodamus/octodamus-core◑ WATCH
·
Nutlope/hallmark◆ SUBSCRIBE
·
NanmiCoder/MediaCrawler◑ WATCH
·
NVIDIA/Model-Optimizer◑ WATCH
·
MinishLab/semble◆ SUBSCRIBE
·
MikkoParkkola/mcp-gateway◑ WATCH
·
Mibayy/token-savior◆ SUBSCRIBE
·
MemPalace/mempalace◑ WATCH
·
MastadoonPrime/sylex-search◑ WATCH
·
Lightning-AI/litgpt◑ WATCH
·
JimLiu/baoyu-skills◑ WATCH
·
·
HelpCode-ai/anythingmcp◑ WATCH
·
Graphify-Labs/graphify◑ WATCH
·
Gitlawb/openclaude◑ WATCH
·
Gentleman-Programming/engram◆ SUBSCRIBE
·
GeeeekExplorer/nano-vllm◑ WATCH
·
GTCC777/pulsenetwork-mcp◑ WATCH
·
GAIR-Lab/Light-MER◑ WATCH
·
EKKOLearnAI/hermes-studio◑ WATCH
·
DeusData/codebase-memory-mcp◆ SUBSCRIBE
·
·
CoplayDev/unity-mcp◑ WATCH
·
ConstantineB6/comfy-pilot◑ WATCH
·
Cifero74/mcp-apple-music◑ WATCH
·
CNSLabs/agreements-api-sdk◑ WATCH
·
BuilderIO/agent-native◇ REFERENCE
·
BluesPrince/thiri-mcp◑ WATCH
·
BerriAI/litellm◑ WATCH
·
·
Arize-ai/phoenix◆ SUBSCRIBE
·
AgentOps-AI/agentops◑ WATCH
·
·
AceDataCloud/SeedreamMCP◑ WATCH
·
AceDataCloud/MCPNanoBanana◑ WATCH
·
AceDataCloud/FluxMCP◑ WATCH
·
AIDataNordic/alexandria-mcp◑ WATCH
·
·
2s-io/sdk◑ WATCH
·
1mcp-app/agent◑ WATCH
Files · browse · copy path · preview
Select a file to preview