The Agent Contract is the foundational governance document for the Decision Rain Library Project. It establishes exactly what the AI assistant is permitted to do, what it is forbidden from doing, and when it must stop and wait for the operator before taking any further action. Everything in the system flows from this contract.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/XxYouDeaDPunKxX/decision-rain-library-project/llms.txt
Use this file to discover all available pages before exploring further.
Core Authority Principle
The operator is the only validation authority in the Decision Rain Library Project. No decision becomes canonical until the operator explicitly approves it. The assistant exists to prepare, surface, and refine proposals — not to make final judgments. Thorough research and confident analysis do not grant the assistant any additional authority to act.What the AI May and May Not Do
| AI MAY | AI MAY NOT | |
|---|---|---|
| Links | Preserve original links, save to 00_INBOX | Move, promote, or archive without operator approval |
| Research | Inspect, research, and gather official and community evidence | Infer substance from title, URL, stars, or marketing copy |
| Analysis | Analyze, compare, and surface conflicts or ambiguities | Treat analysis as a validated decision |
| Proposals | Propose tags, classifications, StackFit, Verdict, Next, and taxonomy gaps | Apply any proposal as final without explicit operator approval |
| Taxonomy | Report tag gaps and propose the smallest new tag | Create new tag families, values, or synonyms |
| SYSTEM | Read and apply approved SYSTEM documents in order | Change SYSTEM documents without operator approval |
| Normalization | Define a plan and sample for review | Execute batch normalization without explicit operator approval |
Default Flow When a Link is Provided
When the operator provides a link, the assistant follows this exact sequence every time — with no exceptions.Save to 00_INBOX
The assistant saves the original link to
00_INBOX with status/pending-review and one next/* tag. This step preserves the trace of the entry’s arrival. It is not a decision and does not validate anything about the link.Research immediately
The assistant immediately researches the link using all relevant available tools. It gathers both official evidence (docs, README, maintainer notes, release notes, pricing, source, examples, tests) and community evidence (issues, discussions, user reports, forums, HN/Reddit when relevant, adoption signals, breakage reports, pricing and setup complaints). Official and community evidence are kept separate.
Prepare a proposal
The assistant assembles a full proposal: tags, collection placement, StackFit, Verdict, Next action, risks, alternatives, missing evidence, and taxonomy gaps. Every proposal item is labeled with a certainty level — verified, inferred, uncertain, or operator-decision-required.
Operator Gate
Every item below requires explicit operator approval before the assistant may act. The operator must state approval clearly — the assistant does not infer approval from context, silence, or prior decisions.- Final collection placement
- Final
status/*tag - Final
truth/*tag - Final StackFit
- Final
next/*tag - Promotion out of
00_INBOX -
20_LIBRARYplacement - Archive (move to
90_ARCHIVE) - Any SYSTEM document change
- Any taxonomy change (new families, values, or synonyms)
- Batch normalization
- Overwriting existing decision notes
Reading Order for SYSTEM Documents
The assistant reads00_AGENT_CONTRACT first, then SYSTEM documents (01 through 05) in order, before classifying or updating any serious entry. This order is mandatory, not advisory.
SYSTEM documents are governance documents — they define the rules of the library. They are not ordinary bookmarks and must never be treated as entries to classify, promote, or archive.