vault-agent / rules
AGENTS.md2 starsChanged 4 months agoMITView on GitHub
- One agent per file, prompt as a sibling .md in prompts/, loaded via prompt_path. A deterministic agent has no prompt and load_prompt() raises if you give it one. - Every model call goes through ForcedToolCaller (llm.py): forced single tool, streaming, cache-controlled system block, the shared retry matrix, usage and trace recorders. Do not call the SDK directly from an agent — grep messages.create src/vault_agent must stay empty (ADR-0010). - The LLM part is injectable. Deterministic core in the agent,…
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.
In this repository2 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- project-docsSkill
The file
We split this agents.md into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
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 14
Do not call the SDK directly from an agent — grep messages.create src/vault_agent must stay empty (ADR-0010).
≈ “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
Agent conventions
- One agent per file, prompt as a sibling
.mdinprompts/, loaded viaprompt_path. A deterministic agent has no prompt andload_prompt()raises if you give it one. - Every model call goes through
ForcedToolCaller(llm.py): forced single tool, streaming, cache-controlled system block, the shared retry matrix, usage and trace recorders. Do not call the SDK directly from an agent —grep messages.create src/vault_agentmust stay empty (ADR-0010). - The LLM part is injectable. Deterministic core in the agent, model call behind a seam the tests replace. The suite runs with no API key.
- A truncated response is a typed failure, not an empty payload:
LLMCallError.truncated. List-shaped output splits viallm.call_with_truncation_splitwith a domain merge; a single coherent artefact has only the budget lever. - Never guess a value the input does not carry. Flag it:
state.flag(...)with aFlagKindand the asset. An honest gap is output; an invented column is a defect. Consumers branch onkind/asset, never on the message text. - Prompt steering is registered, not inlined:
DV_MODELING_RULESentries carry an id, origin and backstop link, and the ledger (docs/architecture/steering-ledger.md) records the evidence. Adding or deleting a rule regeneratestests/fixtures/steering/modeler_rules_pre_wp16.txtin the same commit, with the pre-WP16 block asserted to remain a byte-identical prefix. - DV2.0 knowledge lives in
rules/, not in a prompt and not at a call site. If you need a hub's staging key column, a satellite's feed or its payload relations, a role-qualified column or a normalised identifier, call the helper. - Log at agent boundaries (INFO, with construct counts), payload sizes at DEBUG. The library never configures handlers or levels.
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.

