Tesseract
AGENTS.md4 starsChanged 14 days agoMITView on GitHub
Never import C-extension chains (numpy, torch, sentence_transformers) inside an MCP tool body or anything a tool calls lazily. On Python 3.14 + Windows, the first such import inside the FastMCP worker thread stalls until the next stdin message arrives — the server appears to hang forever on single requests (root-caused 2026-07-11; see the audit session log in the vault). Eager-import at server startup in the main thread instead: server._warm_start() exists for exactly this. Verify with python scripts/probe_server.py.
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
The file
We split this AGENTS.md into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 22
Every pin's sha256 lock must cover each asset its release ships.
- Distinctive: line 27
Never import C-extension chains (numpy, torch, sentence_transformers) inside an MCP tool body or anything a tool calls lazily.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Learned User Preferences
- Executes implementation plans via subagent-driven-development: fresh subagent per task, spec compliance review first, then code quality review.
- Prefers slow, step-by-step deployment walkthroughs when doing something for the first time.
- For Oracle Cloud VM creation: use On-demand capacity (not Dedicated host), disable Confidential computing on the instance, and pair A1 Flex with an aarch64 Ubuntu image (e.g. Ubuntu 24.04 Minimal aarch64).
- Oracle image list "BM Confidential computing" in the Security column does not mean every image requires confidential compute; pick aarch64 Ubuntu for A1 and turn off confidential compute only if the instance form shows an incompatibility error.
- MCP live sync (
python -m tesseract_mcp.mcp_syncwithout--check) and live-vault styling/provision require explicit user consent; agents dry-run with--checkand throwaway vaults only. - Skill sync (
python -m tesseract_mcp.skill_syncwithout--check) to the real~/.claude/skillsrequires explicit user consent; agents run--check(optionally with a scratch--dest) only. - Obsidian screenshots from the live vault need human review of every image before commit (LiveSync banner and personal content).
Learned Workspace Facts
tesseract-mcprepo:C:\Users\Taimoor\Documents\GitHub\tesseract-mcp, base branchmain; run Python with.venv\Scripts\python; tests with.venv\Scripts\python -m pytest -q.- Live Tesseract vault:
C:\Vaults\Tesseract; markdown is source of truth; agent writes quarantined toClaude/. - Sync deployment: Oracle Always Free A1 VM running
tesseract-mcp/server/stack (CouchDB + Caddy + DuckDNS); Obsidian Self-hosted LiveSync on the vault. - Semantic graph markdown in
Claude/Graph/is authoritative; derived cache at~/.tesseract-mcp/graph.db; run--rebuild-onlywhen graph markdown changed without note re-indexing. - Vault installer (
scripts/install_conventions.py) is idempotent and never overwrites existing files. mcp-servers.jsonat repo root is the curated MCP manifest;tesseract_mcp.mcp_syncadditively registers servers into~/.claude.jsonviaclaude mcp add(never modify or remove existing entries).vault-template/plugins.jsonpins Obsidian plugins;provision.pyauto-enables every successfully installed pin ([s.id for s in specs if s.id not in errors]).- On Windows,
mcp_syncresolves theclaudeCLI viashutil.which(injectable in tests) so npm'sclaude.cmdshim is found. - Obsidian plugin pins must use release tags whose manifest version matches the tag (avoids provision idempotency drift and LiveSync churn).
- Every pin's
sha256lock must cover each asset its release ships; a pin bump updates the lock perdocs/PLUGIN_LOCK.md(independently computed digests, cross-checked against GitHub's published digest when present). - (Migrated 2026-07-09 from the retired Sentinal-ESG workspace, where early tesseract work happened.)
MCP server rule: no lazy heavy imports in tool bodies
Never import C-extension chains (numpy, torch, sentence_transformers)
inside an MCP tool body or anything a tool calls lazily. On Python 3.14 +
Windows, the first such import inside the FastMCP worker thread stalls
until the next stdin message arrives — the server appears to hang forever
on single requests (root-caused 2026-07-11; see the audit session log in
the vault). Eager-import at server startup in the main thread instead:
server._warm_start() exists for exactly this. Verify with
python scripts/probe_server.py.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

