[recipes] Session boot sequence with priority-based loading#353
Closed
gleesonb wants to merge 327 commits into
Closed
[recipes] Session boot sequence with priority-based loading#353gleesonb wants to merge 327 commits into
gleesonb wants to merge 327 commits into
Conversation
…ntrib/matthallett1/source-filtering
…ltering - Canonical OB1 MCP server with 4 tools: capture_thought, search_thoughts, list_thoughts, thought_stats - Source filtering on search_thoughts, list_thoughts, thought_stats (optional source param) - Auth via x-brain-key header and ?key= query param against MCP_ACCESS_KEY - deno.json with pinned canonical dependency versions - Update .gitignore to track Edge Function source files while ignoring Supabase temp/config Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replaced custom monkeyrun Edge Function with OB1 canonical MCP server. 4 tools (capture_thought, search_thoughts, list_thoughts, thought_stats), source filtering on 3 tools, auth via query param and header. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Supabase Edge Functions pass the function name as part of the request
path (e.g., /professional-crm/mcp), but all extensions used explicit
routes like app.post("/mcp") which only matches /mcp. This caused every
extension to return 404 when deployed.
Changed to wildcard routing (app.post("*"), app.get("*")) to match
regardless of path prefix, consistent with the core open-brain-mcp
server pattern.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
z.record(z.any()) in the details parameter breaks with Zod 4.x and MCP SDK 1.24.3 (_zod property error). Changed to z.string() accepting a JSON string, which is compatible and functionally equivalent since the JSONB column accepts string input. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Local deployment copies of the 4 extension MCP servers with fixes applied: edge runtime import, wildcard routing for Supabase path handling, and z.record compatibility fix for household-knowledge. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Deployment-ready copies with edge runtime import and wildcard routing fix applied. All 6 OB1 extensions now deployed. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Same wildcard routing fix as the other 4 extensions — Supabase passes the function name in the request path. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
server.tool() wraps params in z.object() internally. Passing a z.object() schema double-wraps it, producing empty parameter schemas. Changed all 8 tool registrations to pass schema.shape instead. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Tool handlers returned raw strings instead of MCP content objects
({ content: [{ type: "text", text: "..." }] }). Added wrap() helper
to all 8 tool registrations with error handling.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
OB1 pattern: users copy from extensions/ and deploy themselves. Removed supabase/functions/ from tracking — these are local deployment copies, not part of the repo standard. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Remove .planning/ files from tracking (already gitignored), fix broken link to nonexistent recurring-tasks primitive, add numbered step list to source-filtering README for review check compliance. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
… guide Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…uide
Three categories of fixes:
1. STEP ORDERING FIX: Moved `supabase secrets set MCP_ACCESS_KEY` from
Step 5 to Step 6 — it previously told users to run a Supabase CLI
command before telling them to install the CLI.
2. CODE BLOCK CLEANUP: Every code block now contains exactly one
copy-pasteable command. Previously, Mac/Linux/Windows commands were
jammed into single blocks with comment headers, forcing users to
visually parse which line to copy. Affected sections:
- Access key generation (Mac/Linux vs Windows separated)
- Supabase CLI install (Mac, Windows Scoop setup, Windows Scoop
bucket, Windows Scoop install, and Linux each get their own block)
- Login and project linking (split into separate steps)
- Secret setting (each secret gets its own block with context)
3. CLICKABLE LINKS: Bare-text URLs converted to markdown links:
supabase.com, openrouter.ai, openrouter.ai/keys,
openrouter.ai/models, chatgpt.com (x2)
Also fixed stale "step-10" placeholder references → "step-5" in three
places (MCP Connection URL, Claude Code header, mcp-remote config).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two sections rewritten for zero-ambiguity copy-paste: 1. ADD DEPENDENCIES: Previously said "Create supabase/functions/ open-brain-mcp/deno.json:" followed by a JSON block — unclear to beginners who've never created a file from a terminal. Now provides a single terminal command per OS (Mac/Linux heredoc, Windows PowerShell here-string) that creates the file directly. 2. WRITE THE SERVER: Previously linked to an external guide for the MCP server code, requiring users to navigate away, find the right block, copy it, figure out how to open index.ts, and paste. Now includes the full server code inline as a terminal command per OS — one copy, one paste, file is written. No text editor needed. Both sections follow the pattern established by the rest of the guide: one code block = one action = one copy-paste. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Added 🟩 to all Mac/Linux labels and 🟦 to all Windows labels throughout the guide for quick visual scanning (9 labels total) - Collapsed all three SQL queries in Step 2 (thoughts table, search function, RLS policy) into <details> blocks - Server code and deno.json OS-specific blocks were already collapsed from the previous commit
Major rewrite of the MCP server deployment flow: 1. NEW: "Create a Project Folder" section teaches users to create a folder in Finder/Explorer, copy the path, and cd into it. Bridges the GUI they know into the terminal they're learning. 2. NEW: "Initialize and Link" adds `supabase init` before linking, establishing the project structure in their chosen folder. 3. NEW: "Download the Server Files" replaces ~750 lines of heredoc blocks with curl/Invoke-WebRequest commands that pull index.ts and deno.json directly from the repo's new server/ directory. 4. NEW: server/index.ts and server/deno.json added to the repo as the canonical source files for the core MCP server. File reduced from 1441 to 642 lines. Every code block is still one copy-paste action, but users no longer create files manually. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Three-part change to formalize the curl-from-GitHub deployment pattern:
1. PRIMITIVE REWRITE: primitives/deploy-edge-function/ now uses curl/
Invoke-WebRequest to download server files directly from the repo.
Includes project folder reminders, OS-specific commands with color
indicators, and guidance on reusing access keys across extensions.
2. EXTENSION TABLES: All 6 extension READMEs updated from
"Server code | This extension's index.ts" to
"Download path | extensions/{name}" — consumed by the primitive
to construct the download URLs automatically.
3. AGENT SPEC: New extensions/_template/AGENT_SPEC.md provides a
machine-readable specification for generating new extensions. Covers
all 5 required files (deno.json, metadata.json, schema.sql,
index.ts, README.md) with exact patterns, naming conventions,
validation checklist, and example prompt. Designed so an AI agent
can produce a complete, correctly-formatted extension in one pass.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Full formatting rewrite for scannability and visual hierarchy: - Shield badges (for-the-badge style) for all 8 main step headers - Numbered sub-steps (6.1, 6.2, etc.) replacing unnamed ### headings - GitHub alert callouts replacing plain blockquotes: > [!IMPORTANT] for must-do actions > [!TIP] for helpful context > [!WARNING] for security/data concerns > [!CAUTION] for easy-to-miss gotchas > [!NOTE] for supplementary info - Emoji markers: ✅ Done when (step completion), ❌ (troubleshooting), 🔖 (credentials to save), 🔗 (links/references), 📋 (SQL blocks) - Section emoji headers: ❓ Troubleshooting, 🔍 Under the Hood, ➕ Optional, 🎉 What You Built, ➡️ Next Step - Collapsible SQL blocks now labeled "(click to expand)" No content changes — same guide, dramatically easier to scan. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Step 6 was a visual nightmare — OS-specific commands interleaved throughout made it easy to grab the wrong block (confirmed by testing). Now Step 6 has two collapsible sections: "Step 6 — Mac / Linux" and "Step 6 — Windows". Each contains the complete flow for that OS with no cross-references. Users click their OS once and follow a clean, linear path: 6.1 Create a Project Folder (OS-specific instructions) 6.2 Install the Supabase CLI (brew vs scoop) 6.3 Log In 6.4 Initialize and Link 6.5 Set Your Secrets 6.6 Download the Server Files (curl vs Invoke-WebRequest) 6.7 Deploy Shared steps (login, init, secrets, deploy) are duplicated in both sections — small cost for zero ambiguity. Also added a warning to Step 5's key generation reminding users to copy only the command for their OS. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Each AI client (Claude Desktop, ChatGPT, Claude Code, Other Clients) is now in its own collapsible <details> block. Users click the client they use and see only those instructions — no scrolling past irrelevant setup for other clients. Callout blocks (NOTE, WARNING, CAUTION, TIP) nested inside the collapsible sections for inline context. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Troubleshooting, How It Works, Optional Capture Sources, and What You Built are now collapsible <details> blocks. Keeps the post-setup content out of the way — users who need it can expand it, everyone else sees a clean ending with the Next Step section visible. "Your Next Step" stays visible (not collapsed) since it's the main call to action after completing the guide. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Each step badge now walks the spectrum — grey number label with a colored title background that progresses through the rainbow: Step 1: Red (#E53935) Step 2: Deep Orange (#F4511E) Step 3: Orange (#FB8C00) Step 4: Green (#43A047) Step 5: Teal (#00897B) Step 6: Blue (#1E88E5) Step 7: Indigo (#5C6BC0) Step 8: Purple (#8E24AA) Gives instant visual progress cues — you can tell where you are in the guide at a glance by the color alone. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Removed the copy-paste text block and old .docx file. Replaced with a prominent CAUTION callout and download link for the new .xlsx credential tracker spreadsheet. The tracker is now impossible to miss — it's a giant warning block that tells users to stop and download before proceeding. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
🛑🛑🛑 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Removed: If You Get Stuck, How This Works, After You're Done, and Cost Breakdown sections. All redundant with content already at the end of the guide. People arriving here know what they want. Reframed "What You're Building" as the core foundation that everything else in the repo builds on — not a standalone project description. Two lines, two services, go. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Straight to the point. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Visual guide showing exactly where the download button is on GitHub. Because if we're going to make a huge deal out of downloading it, we should show them how. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…lanshurafa/entity-wiki [recipes] Entity wiki pages from knowledge graph
…lanshurafa/wiki-synthesis [recipes] Wiki synthesis + autobiography pipeline
…odex/wiki-compiler [recipes] Add wiki compiler orchestration recipe
…odex/chatgpt-mcp-compat Improve ChatGPT MCP compatibility
- Fixed edge classifier source-aware sampling (0 → 3290 candidate pairs) - Fixed edge classifier RPC fallback for parameter mismatch - Added gpt-4o-mini to PRICING table - Rewrote n8n Substack workflow (direct Supabase REST) - Added n8n Gmail daily ingestion workflow - Added n8n DOK refinement loop workflow - Daily digest script, AI memory import script - Server index.ts: OpenAI direct (replaces OpenRouter) - Updated .gitignore to exclude exports/backup/compiled-wiki Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…nclaw-launch' # Conflicts: # .gitignore # dashboards/open-brain-dashboard-next/app/thoughts/[id]/page.tsx # dashboards/open-brain-dashboard-next/components/KanbanBoard.tsx # dashboards/open-brain-dashboard-next/lib/api.ts # dashboards/open-brain-dashboard-next/lib/types.ts
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds a P1-P4 priority loading system for AI agent memory boot sequences. P1 rules load every session, P2 reference on-demand, P3 runbooks before touching systems, P4 integrations only when needed. Includes /refresh pattern for context drift recovery in long sessions. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
github-actions
Bot
added
dashboard
|
Hey @gleesonb — welcome to Open Brain Source! 👋 Thanks for submitting your first PR. The automated review will run shortly and check things like metadata, folder structure, and README completeness. If anything needs fixing, the review comment will tell you exactly what. Once the automated checks pass, a human admin will review for quality and clarity. Expect a response within a few days. If you have questions, check out CONTRIBUTING.md or open an issue. |
|
Closing — this was meant for an internal project, not the public repo. Sorry for the noise. |
1 similar comment
Author
|
Closing — this was meant for an internal project, not the public repo. Sorry for the noise. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What it does
Adds a priority-based memory loading system (P1-P4) that configures AI agents to load the right context at the right time during session boot. Includes a
/refreshpattern for context drift recovery in long sessions.What it requires
Tested on my own instance
Yes — I use this pattern daily with Claude Code and Open Brain.
Pattern attribution
Priority-based loading pattern adapted from open-brain-template by Tony Finley.
🤖 Generated with Claude Code