# Final Decision > use once a candidate has reached HUMAN_REVIEW (or you want an interim standing check at any point via hunt rank) to compile the full evidence-based report and present it to the human for their approve/reject decision. Never sets QUALIFIED itself. - Skill: `zaidadaqqa/final-decision` (Agent Skill) - Install (CLI): `npx skillmds@latest add zaidadaqqa/final-decision` - Raw SKILL.md: https://api.skillmd.com/api/skills/zaidadaqqa/final-decision/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Docs & Writing - Author: zaidadaqqa (https://skillmd.com/u/zaidadaqqa) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/zaidadaqqa/final-decision --- # Final Decision Compiles the complete, evidence-based standing of a candidate and presents it to the human so **they** can approve or reject it. This Skill drafts nothing on the human's behalf and never implies its own summary or recommendation IS the decision. Full explanation of what each status means lives in `references/decision-framework.md`; the exact gate/state/ranking mechanics live in `references/decision-gates.md`. Read both before using this Skill. ## When to use this - Once a candidate reaches `HUMAN_REVIEW` (the natural end of the happy-path chain — see `decision-gates.md` SM-04). - Or at any earlier point, as an interim standing check, via `hunt rank` alone — this is safe to run repeatedly and has no side effects beyond recording the recomputed `final_status` (`decision-framework.md`). ## Step 1 — Get current standing ``` hunt rank ``` This re-applies hard gates and Pareto ranking using only evidence already collected (fetches nothing new). It sets `final_status` to `REJECTED` (gate fired), `NEEDS_MORE_DATA` (gates passed, zero ranking dimensions computed), or `PROMISING` (gates passed, Pareto-ranked) — **never** `QUALIFIED`. Always re-run this rather than relying on a standing computed earlier in the conversation; new evidence may have changed it. ## Step 2 — Get the full evidence report ``` hunt show ``` This is the full evidence-first report. Every section without data says so explicitly (`NOT_YET_ANALYZED`, `NO_DATA`, `NOT_PROVIDED`, etc.) — never a silent omission (`REP-01`). Use this, not memory of earlier conversation turns, as the source of truth for what's actually recorded. ## Step 3 — Present everything, organized and complete Present to the human, in full: - **Which gates ran and their result** (per `decision-gates.md` GATE-01/02) — and if any fired, exactly which `CONFIRMED_*` value fired it. Never imply a gate fired on a `SUSPECTED_*`/`PENDING_*`/`UNKNOWN` value — that is structurally impossible. - **The four ranking dimensions** (`decision-gates.md` RANK-03) — which are present, their actual values, and which are omitted because that stage hasn't run yet. Never treat an omitted dimension as if it scored zero. - **Demand** — from `hunt demand-show`/report data, including the qualification and its advisory (non-enforced) status. - **Competition** — competitor autopsy findings from `hunt show`. - **VOC / customer pain points** — from review analysis and/or manually recorded pain points, including severity distribution. - **Differentiation** — the copycat-resistance classification if the stage has run, `NOT_YET_ANALYZED` if not. - **Red team** — `EVIDENCED` vs. `HYPOTHESIS` findings, distinctly labeled. - **Economics** — the full base case and stress-test results from `hunt economics-show`, or `NOT_PROVIDED` if `economics-set` was never called (`economics.md` ECON-11). Never compute or restate these numbers differently than what the CLI printed. - **Supplier/IP status** — the recorded `supplier_status`/`ip_status` (default `UNKNOWN` if never set), and whether either is human-confirmed. - **What's still missing** — explicitly list every stage/field not yet recorded, so the human can see the gaps as clearly as the findings. ## Step 4 — Hand the decision to the human, explicitly State the current `final_status` (`PROMISING`/`REJECTED`/ `NEEDS_MORE_DATA`) and what it actually means per `decision-framework.md` — in particular, if `PROMISING`, say plainly that this is a gate-pass technicality, not a business verdict, and summarize exactly which dimensions support it. Then present the decision command for the human to run themselves: ``` hunt decide approve --reason "" hunt decide reject --reason "" ``` This Skill may summarize tradeoffs if asked, but must not draft the `--reason` text as if it were the human's own words, must not run `hunt decide` itself under any circumstance, and must not phrase its own summary in a way that could be mistaken for the decision itself having already been made. ## What this Skill must never do - Never call `hunt decide` itself, for either `approve` or `reject`. - Never present a `PROMISING` result as if it were a positive verdict on the product's viability. - Never imply that a `REJECTED` gate result reflects human judgment when it was actually automatic (or vice versa) — always be clear which is which (`decision-framework.md`). - Never omit a "still missing" section — an incomplete picture presented as if it were complete is itself a form of hallucination by omission. See `references/decision-framework.md` for the full explanation of what each status means and the relationship between `hunt rank` and `hunt decide`, and `references/decision-gates.md` for the exact gate/state/ ranking mechanics and rule-preservation table.