Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/blackwell-systems/knowing/claude-mdgit clone --depth 1 https://github.com/blackwell-systems/knowingWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.06413 | $0.06413 |
| Opus 5 | $0.03207 | $0.03207 |
| Sonnet 5 | $0.01283 | $0.01283 |
| Haiku 4.5 | $0.00641 | $0.00641 |
Grade A, and why
knowing CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 290 lines — stays where its author put it; the contents beside it link to each section on GitHub.
knowing
Self-adapting code intelligence engine. Single Go binary, zero runtime deps. Gets smarter with scale, not dumber: observes its own graph density and adjusts retrieval strategy automatically. 38 edge types, 23 extractors, 28 MCP tools.
Build & Test
GOWORK=off go build ./... # build (GOWORK=off required: go.work refs missing module)
GOWORK=off go test ./internal/... # unit tests
GOWORK=off go test ./cmd/... # CLI tests
GOWORK=off go test ./bench/... # benchmark harnesses (some need pre-indexed repos)
Benchmark (P@10 evaluation)
# Full corpus, sequential (official numbers, ~20 min with pre-embedded vectors)
BENCH_EMBEDDINGS=1 BENCH_ADAPTERS=knowing GOWORK=off go test ./bench/cross-system/ -run TestCrossSystem -v -timeout 0
# Full corpus, parallel (iteration mode, ~5 min, P@10 ~0.022 lower due to ONNX CPU contention)
BENCH_PARALLEL=1 BENCH_EMBEDDINGS=1 BENCH_ADAPTERS=knowing GOWORK=off go test ./bench/cross-system/ -run TestCrossSystem -v -timeout 0
# Single repo (fast iteration, no embeddings)
BENCH_REPOS=django BENCH_ADAPTERS=knowing GOWORK=off go test ./bench/cross-system/ -run TestCrossSystem -v -timeout 10m
# With embeddings on single repo (~2 min with pre-embedded vectors)
BENCH_EMBEDDINGS=1 BENCH_REPOS=django BENCH_ADAPTERS=knowing GOWORK=off go test ./bench/cross-system/ -run TestCrossSystem -v -timeout 30m
# Pre-embed all nodes (one-time, ~2 hours for full corpus, skips phantoms)
knowing enrich embeddings -db <repo>/.knowing/graph.db
# Diagnostic env vars (compose freely, no reindex needed):
BENCH_EXCLUDE_EDGES=similar_to,type_hint_of # exclude edge types from RWR walk
BENCH_BFS_DEPTH=2 # limit walk depth (default 4)
BENCH_PREFER_TYPE_SEEDS=1 # force type-seed preference
BENCH_HUB_DAMPEN=50 # penalize nodes with in-degree >50
BENCH_RERANK_WEIGHT=0.5 # blend original + embedding scores
BENCH_COHERENCE_BONUS=0.3 # file-based packing coherence
BENCH_MAX_SEEDS=25 # override max seed count
BENCH_ADAPTIVE_SEEDS=1 # enable adaptive seed count
BENCH_GAP_THRESHOLD=5 # gap-fill activation threshold (default 5)
BENCH_PARALLEL=1 # parallel repo execution (fast, ~0.022 P@10 lower)
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 290 lines · 6,413 tokens per session scan A ecee88680c65
knowing CLAUDE.md is an instructions file published in the GitHub repository blackwell-systems/knowing (18 stars, last pushed 1mo ago), licensed MIT. It adds 6,413 tokens to every session, about $0.0321 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
trace-mcp CLAUDE.md
Instructions for nikolai-vysotskyi/trace-mcp, covering trace-mcp development guide, what this project is, agent behavior — read before every task, no flattery, no filler and disagree when the premise is wrong.
trace-mcp AGENTS.md
Instructions for nikolai-vysotskyi/trace-mcp: IMPORTANT: For ANY code exploration task, ALWAYS use trace-mcp tools first. NEVER use Read/Grep/Glob/Bash(ls,find) for navigating source code.
m1nd AGENTS.md
Instructions for maxkle1nz/m1nd, covering agents.md — working guide for any coding agent on m1nd, the gates (must pass — these are the ci), git identity — absolute, no-leak — reputation rule (public repo) and how work lands — bursts, not pr-per-fix.
codeGraph-branch CLAUDE.md
Instructions for lagrahhn/codeGraph-branch, covering claude.md, project overview, build, test, run, single test file / pattern and architecture.
mcp-server-zig CLAUDE.md
Instructions for sadopc/mcp-server-zig, covering claude.md, build & run, architecture, tools and key patterns.
ummon CLAUDE.md
Instructions for Nayshins/ummon, covering ummon development guide, build & test commands, build the project, run the project and run with specific command.