What problem does it solve? Teams running multi-agent orchestration pipelines cannot know what a task will cost in tokens, compute time, or dollars until after it runs. This Skill provides a methodology and format contract for producing pre-execution estimates as honest ranges with disclosed confidence, eliminating false-precision point guesses. ## Core Features & Use Cases - Range-based estimation: Produces token and agent-compute-time estimates as { low, high } ranges with per-axis confidence, never point values. - Actuals-grounded cost figures: Derives dollar costs only from observed observability/costs/ snapshots via a tier-to-model binding table, omitting cost with explicit disclosure when no usable snapshot exists β never falling back to list prices. - Four-part disclosure contract: Every estimate record carries Included, Excluded, Confidence rationale, and Failure direction sections, with a validation checklist that rejects verdict or go/no-go language. - Calibration loop: Refines per-stage token ranges against accumulated per-PR actuals records from the repo's own history. - Use Case: Before committing to a spec implementation, ask for an estimate and receive "10kβ18k tokens, medium confidence, cost omitted β no snapshot exists yet" so you can make an informed decision. ## Quick Start Ask the agent to estimate the token usage, compute time, and cost of a task or spec before running it, and review the returned range-based estimate record with its disclosure sections.
