agent-estimate 0.2.0

Effort estimation for AI coding agents — PERT + METR + wave planning

agent-estimate

agent-estimate is a CLI for estimating delivery time of AI-agent work using:

• three-point PERT estimates
• METR-style model reliability thresholds
• dependency-aware wave planning
• explicit review overhead modes (none, standard, complex)
• non-coding task type estimation (brainstorm, research, config, docs)
• multi-agent session estimation

Installation

Install from PyPI:

pip install agent-estimate

Install from source for development:

python -m venv .venv
source .venv/bin/activate
pip install -e '.[dev]'

Quick Start

Estimate one task from the command line:

agent-estimate estimate "Implement OAuth login flow"

Show version:

agent-estimate --version

Claude Code Plugin

agent-estimate includes a Claude Code plugin for interactive estimation in Claude Code sessions.

Install

Option 1 — From marketplace:

/plugin marketplace add haoranc/agent-estimate
/plugin install agent-estimate@agent-estimate-marketplace

Option 2 — Local development:

claude --plugin-dir /path/to/agent-estimate

Prerequisite: The CLI must be installed first: pip install agent-estimate

Plugin Usage

/estimate Add a login page with OAuth
/estimate --file spec.md
/estimate --issues 1,2,3 --repo myorg/myrepo
/validate-estimate observation.yaml
/calibrate

Codex Skill Layout

For Codex-oriented tooling, this repo includes a Codex-specific skill at:

• .agent/skills/estimate/SKILL.md

The Claude plugin skill remains at:

• skills/estimate/SKILL.md

Both skills cover the same CLI capabilities (estimate, validate, calibrate) but are phrased for their respective ecosystems.

Usage Examples

Estimate tasks from a text file:

agent-estimate estimate --file tests/fixtures/tasks_multi.txt

Output JSON for downstream tooling:

agent-estimate estimate "Refactor auth pipeline" --format json

Estimate directly from GitHub issues:

agent-estimate estimate --repo haoranc/agent-estimate --issues 11,12,14

Validate estimate vs observed outcome and persist to calibration DB:

agent-estimate validate tests/fixtures/observation_valid.yaml --db ~/.agent-estimate/calibration.db

TestPyPI Validation

Manual local publish (requires TestPyPI API token configured for twine):

python -m build
python -m twine check dist/*
python -m twine upload --repository testpypi dist/*
pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple agent-estimate

Or run the GitHub Actions workflow TestPyPI Dry Run to publish and smoke-test install end-to-end.

Default METR Thresholds

The default model thresholds are defined in src/agent_estimate/metr_thresholds.yaml:

Model	p80 threshold
Opus	90 minutes
GPT-5.3	60 minutes
GPT-5	50 minutes
GPT-5.2	55 minutes
Gemini 3 Pro	45 minutes
Sonnet	30 minutes

Agent Config Example

Pass a custom config file with --config:

agents:
  - name: Claude
    capabilities: [planning, implementation, review]
    parallelism: 2
    cost_per_turn: 0.12
    model_tier: frontier
  - name: Codex
    capabilities: [implementation, debugging, testing]
    parallelism: 3
    cost_per_turn: 0.08
    model_tier: production
settings:
  friction_multiplier: 1.15
  inter_wave_overhead: 0.25
  review_overhead: 0.2
  metr_fallback_threshold: 45.0

Then run:

agent-estimate estimate "Ship packaging flow" --config ./my_agents.yaml

Contributing

1. Fork and create a branch from main.
2. Install dev dependencies:

   pip install -e '.[dev]'
   

3. Run checks:

   ruff check .
   pytest -q
   

4. Open a pull request with a clear summary and test evidence.

License

MIT