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/novasplace/csm/agents-mdgit clone --depth 1 https://github.com/NovasPlace/CSMWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/novasplace/csm/agents-md)<a href="https://agentmods.dev/instructions/novasplace/csm/agents-md"><img src="https://agentmods.dev/badge/instructions/novasplace/csm/agents-md.svg" alt="Measured on agentmods" height="20"></a>What 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 | $0.02553 | $0.02553 |
| Opus 5 | $0.01277 | $0.01277 |
| Sonnet 5 | $0.00511 | $0.00511 |
| Haiku 4.5 | $0.00255 | $0.00255 |
Grade A, and why
CSM AGENTS.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 4d 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 — 104 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Goal
- SQLite MVP complete (Phase 3). Lint debt reduction complete: Phase L1+L2, L3.1-L3.5, L4-A through L4-K done. Baseline locked at 7 warnings (all in opentui.d.ts, skipped by design).
- Phase 4 (Living State Layer) complete: experience packets, self-model, belief knowledge, advisory context-brief injection. All 4F-C requirements verified.
- Phase 9B (Onboarding Quality + Telemetry) complete: context injection telemetry schema, compaction telemetry audit, observation window active.
- Phase 9C (Database-Wide Compaction Observability) implemented: project/client/runtime attribution, classified failures, cross-session coverage, gross/injection/net token accounting, and safe partial cache-write handling. Production migration/observation pending runtime restart.
- Capability promotion closure: all 7 criteria implemented, independently reviewed, cross-database verified. Unblocked.
- Observation window active — pre-9C compaction baseline captured 2026-07-21: 390 rows from 1 session, 232 skipped, 158 failed, 0 compressed, 4,107,005→4,107,005 estimated tokens, 0 verified savings. Next observation begins after runtime restart applies the attribution migration.
Constraints & Preferences
- Each sub-phase is behavior-preserving, boring, verbatim moves first
- CSM_EMBEDDING_PROVIDER (ollama|openai), OPENAI_API_KEY, OLLAMA_HOST
- Database URL: dev/test=localhost, production=explicit flag
- CI with Postgres service container
- ESLint rules start as warnings, tighten later
- Lint warning baseline: 7 warnings (max-warnings=7 prevents unbounded growth)
caughtErrorsIgnorePattern: '^_'added to@typescript-eslint/no-unused-vars— catch blocks with_errare allowedbetter-sqlite3doesn't support?NNNformat with spread.run()— must use anonymous?parameters- SQLite schema: TEXT for timestamps/JSON/arrays/embeddings; INTEGER PRIMARY KEY AUTOINCREMENT for PKs
- PostgreSQL remains default; SQLite is adapter path, not rewrite; no vector search in SQLite MVP
memories.session_idis nullable (FK on sessions, NULL bypasses it)
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.
- 4d ago First seen · 104 lines · 2,553 tokens per session scan A 35f1d93328ef
CSM AGENTS.md is an instructions file published in the GitHub repository NovasPlace/CSM (41 stars, last pushed 10d ago), licensed MIT. It adds 2,553 tokens to every session, about $0.0128 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
supabase copilot-instructions.md
Copilot instructions for supabase/supabase, covering copilot code review instructions, review policy — read this first, confidence threshold, what not to comment on and what to comment on (priority order).
prest copilot-instructions.md
Copilot instructions for prest/prest, covering copilot instructions for prest go backend, project scope, core engineering principles, solid principles for go and s — single responsibility.
graphjin AGENTS.md
AGENTS.md instructions for dosco/graphjin, covering graphjin agent guide, architectural overview, directory structure & responsibilities, build commands and coding guidelines.
PgQue CLAUDE.md
Instructions for NikolayS/PgQue, covering claude.md -- pgque, project, naming convention, architecture and stack.
blog CLAUDE.md
Claude Code instructions for digoal/blog, covering claude.md, repository overview, repository structure, file naming convention and blog post format.
windmill AGENTS.md
AGENTS.md instructions for windmill-labs/windmill, covering windmill, workflow, documentation, dev environment and per-worktree ports and database.