Tranfu Project Scoring

Tranfu Project Scoring reviews AI workflow projects before validation, internal investment, reusable skill development, public case development, or co-creation. It behaves like a restrained project approval interviewer: it collects core facts first, then produces a decision memo with weighted scoring, hard-gate checks, risks, and one next action.

When to Use

Use this skill when a teammate or agent needs to decide whether an AI workflow idea should be approved, validated, deferred, rejected, or reshaped. It fits company-internal AI initiatives, Tranfu skill assets, public demos, research probes, and external product/MVP ideas.

Do not use it as a generic product backlog sorter. RICE, ICE, and WSJF are better for ranking many already-approved features. This skill is for deciding whether an AI workflow project deserves investment at all.

Expected Output

When input is thin, the skill returns clarification questions instead of a fake score. When enough facts are available, it produces a Markdown decision memo covering project type and weight profile, total score, confidence, information completeness, hard-gate checks, a 10-dimension scoring table, missing information and evidence levels, the riskiest assumption, a 7-day validation experiment, failure preview, and exactly one primary next action. For agent-to-agent handoff, it returns the JSON shape described in references/output-schema.md.

Local Pre-scorer

scripts/score_project.py is a deterministic, conservative Python helper for local checks and regression tests, using only the standard library. It reads JSON from a file path or stdin, returns type: clarification for thin inputs by default, supports --force-score for provisional scoring and --format markdown for readable reports, builds an evidence ledger, adds subcriteria breakdowns, and applies context-specific weights. Tests use pytest on tests/test_score_project.py.

Source of Truth

SKILL.md covers invocation rules, the clarification gate, workflow, and output policy. Framework references live under references/ — scoring-framework.md for dimensions and gates, scoring-contexts.md for weight profiles, scoring-anchors.md for calibration, output-schema.md for output contracts, prompt.md for the reusable evaluator prompt, and examples.md for calibrated examples.