Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.10et.ai/llms.txt

Use this file to discover all available pages before exploring further.

Generate decomposed eval scripts and agent TOMLs from a spec, then run the Karpathy loop until the agent hits 100%.

​
Usage

# Generate from a spec file
tenet build --spec knowledge/MY_SPEC.md --name my-feature

# Generate from inline description
tenet build --name auth --files src/lib/auth.ts --desc "Create auth module with login/logout"

# List all build agents and scores
tenet build --list

# Run an existing build agent
tenet build --run my-feature

# Generate without running
tenet build --spec knowledge/SPEC.md --name my-feature --dry-run

​
Options

FlagDescription
--spec <file>Markdown spec file to generate eval from
--name <name>Agent name (kebab-case)
--files <files...>Target files the agent should create
--desc <text>Inline description of what to build
--listShow all build agents with their latest scores
--run <name>Run an existing build agent (PP autoresearch loop)
-r, --rounds <n>Max rounds when running (default: 5)
--dry-runCreate eval + TOML files but don’t start the agent

​
What It Generates

​
Eval Script (eval/build/<name>.ts)

A TypeScript file with decomposed binary checks: 
const checks = [
  { name: "file-exists", pass: existsSync(resolve("src/lib/auth.ts")) },
  { name: "exports-login", pass: fileContains("src/lib/auth.ts", "login") },
  { name: "type-Session", pass: fileContains("src/lib/auth.ts", "interface Session") },
  { name: "purpose-header", pass: fileContains("src/lib/auth.ts", "@purpose") },
  { name: "compiles", pass: tscPasses() },
]
// Score = passed / total (0.0 → 1.0)

​
Agent TOML (.tenet/agents/build-<name>.toml)

[agent]
name = "build-auth"
scope = "build"
metric = "spec_compliance"
direction = "maximize"
time_budget_seconds = 600

[eval]
script = "eval/build/auth.ts"

[task]
description = "Create src/lib/auth.ts with login(), logout()..."

​
How It Works

  1. Parse — extracts file paths, exports, interfaces from spec markdown
  2. Generate eval — each spec requirement becomes a binary check
  3. Generate TOML — agent config with scope, constraints, task
  4. Run — PP autoresearch loop: try → eval → keep/revert → repeat
  5. Converge — agent iterates until score = 1.0, then creates PR
The build supervisor monitors progress and injects hints when agents stall, hit filename mismatches, or repeatedly fail the same checks.

​
See Also