AGENT SKILL

ramza-discover

Open-ended requirements & stakeholder/goal elicitation BEFORE CLARIFY. Use when the GOAL itself is underspecified (IDEA/STRATEGIC intent) — distinct from CLARIFY, which disambiguates an already-known goal. Read-only, bounded, mechanically countable coverage gate; produces an elicitation summary, never a plan.

License 许可
Apache-2.0
所属仓库
Rynaro/magicite
最近更新
2026-10-14
Stars
★ 2
Forks
0

01 前言 Front Matter

name
ramza-discover
metadata
{ "methodology": "RAMZA" }
description
Open-ended requirements & stakeholder/goal elicitation BEFORE CLARIFY. Use when the GOAL itself is underspecified (IDEA/STRATEGIC intent) — distinct from CLARIFY, which disambiguates an already-known goal. Read-only, bounded, mechanically countable coverage gate; produces an elicitation summary, never a plan.

03 SKILL.md 正文

RAMZA — DISCOVER Skill (pre-CLARIFY elicitation)

Use this skill when the request's objective itself is unknown or latent — not merely when the spec details are ambiguous. DISCOVER runs BEFORE CLARIFY and hands CLARIFY a structured elicitation summary.

When to run DISCOVER (vs CLARIFY)

DISCOVER CLARIFY
Precondition Goal is unknown / latent Goal is known; spec details ambiguous
Intent types IDEA, STRATEGIC, any under-GOALED request REQUEST, CHANGE, BUG_SPEC
Question style Open-ended discovery of goals, stakeholders, success ≤3 plan-shape disambiguation questions
Output Elicitation summary → handed to CLARIFY WHO/WHAT/WHY/CONSTRAINTS parse → Scope

Boundary rule: DISCOVER = "what are we even trying to achieve, and for whom?" CLARIFY = "the goal is clear; which of these plan shapes do you want?" Do NOT run DISCOVER on a well-GOALED request — go straight to CLARIFY.

Why this exists: specification/system-design is the dominant multi-agent failure category (MAST — ~43.8% of failures), and multi-agent systems collapse toward ~30% accuracy when latent stakeholder knowledge is never actively elicited (HiddenBench). CLARIFY alone cannot close this — its ≤3 plan-shape contract assumes the goal is already known. This sub-mode is inherited unchanged from SPECTRA (see DESIGN-RATIONALE.md D2 — the discipline spine is the part the audits praised; DISCOVER's failure mode was never its elicitation design).

Elicitation protocol (bounded, read-only)

Surface, do not assume. For each axis, record what is known and emit a [GAP] marker for each unknown rather than inventing an answer:

  1. Stakeholders — Who requests this? Who is affected? Who approves/reviews? Map the approval chain. [GAP] for each unidentified party.
  2. Latent goals — What is the underlying outcome (not the surface ask)? What is the "job to be done"? Distinguish the stated request from the real goal.
  3. Success metrics — How will we know it worked? Target metric + current baseline (e.g. "p95 < 200ms, currently 800ms"). [GAP] when no measurable criterion exists.
  4. Hard constraints — Budget, deadline, tech-stack lock-in, compliance, platform.
  5. Non-goals — What is explicitly OUT of scope? Surface these early to prevent scope creep downstream.

Bound (no unbounded interview loop)

DISCOVER is elicitation + synthesis, NOT an interactive multi-turn interview agent.

  • Produce ONE elicitation summary in a single pass from available context + the host conversation. Do not loop.
  • If coverage is low (≥2 of the 5 axes are [GAP] with no resolution path), escalate to the human with the gap list — do not fabricate goals to proceed.
  • DISCOVER NEVER produces a plan and NEVER writes code. Its sole output is the elicitation summary handed to CLARIFY.

Mechanical coverage contract

The 5 axes are line items a host can count mechanically — no re-reading DISCOVER's reasoning required:

  • Each axis line in the summary below is exactly one of two states: resolved (the line contains no [GAP] token) or unresolved (the line contains the literal token [GAP]).
  • unresolved_count = the number of axis lines (Stakeholders / Latent goal / Success metrics / Hard constraints / Non-goals — the first 5 checklist lines only, not the Open gaps / Coverage lines that restate them) containing [GAP].
  • Contract: unresolved_count >= 2 ⇒ ESCALATE to human, do not hand off to CLARIFY. unresolved_count <= 1 ⇒ hand off to CLARIFY normally.
  • This is a mechanical count (e.g. grep -c '\[GAP\]' over the 5 checklist lines), not a model self-assessment — a host or lint step can enforce the ≥2-unresolved rule without re-deriving DISCOVER's judgment, exactly the same posture as ramza-lint's grep-level structural checks.

Output: elicitation summary

A compact structured checklist (lives under .spectra/ like all RAMZA output). Each of the first 5 lines is the mechanically-countable unit described above:

## DISCOVER — Elicitation Summary
- [ ] Stakeholders: <list | [GAP]>
- [ ] Latent goal: <restated outcome | [GAP]>
- [ ] Success metrics: <metric + baseline | [GAP]>
- [ ] Hard constraints: <list>
- [ ] Non-goals: <list>
- Open gaps: [GAP] <each unknown, repeated from above for visibility>
- Coverage: <n/5 axes resolved> → handing to CLARIFY | ESCALATE to human

Check the box (- [x]) for a resolved axis (no [GAP] on that line); leave it unchecked (- [ ]) when the line still carries [GAP]. unresolved_count is the number of unchecked boxes, and MUST match the coverage line's arithmetic (5 - n = unresolved_count).

Hard constraints (P0)

  1. READ-ONLY. No code, no file edits, no mutations.
  2. Bounded — single-pass synthesis, [GAP]-and-escalate, never an interview loop.
  3. Emit [GAP] for every unknown; never assume latent intent.
  4. DISCOVER hands its summary to CLARIFY; it does not itself produce a plan.
  5. Coverage is mechanically countable: ≥2 unresolved [GAP] axes ⇒ escalate, computed from the checklist, never eyeballed.
  6. Every output path lives under .spectra/ (see SPEC.md "Output discipline").

See SPEC.md "## Pre-phases" for the full methodology section.


RAMZA — DISCOVER sub-mode

04 同仓库更多 Skills

Rynaro
Rynaro/magicite

wine-dxvk-cache-clear

Clear a stale DXVK shader cache that is causing graphical corruption in a Wine or Proton game. Use when: a game shows corrupted textures or shader flicker that started after a driver or DXVK version change.

★ 2
Rynaro
Rynaro/magicite

steam-download-region-fix

Fix a Steam client stuck at very low download speed by changing its download region. Use when: Steam downloads are far below the local internet speed for every game, not just one.

★ 2
Rynaro
Rynaro/magicite

proton-battleye-eac-toggle

Toggle BattlEye or Easy Anti-Cheat runtime support for a Steam Proton game. Use when: a multiplayer game refuses to launch with an anti-cheat error under Proton.

★ 2
Rynaro
Rynaro/magicite

vivi-verify-incoming

Load when reading any upstream artefact handed off to Vivi that carries a sibling .envelope.json. BLOCKING per ECL §6.2.2: the orchestrator MUST have verified the envelope's SHA-256 (eidolons verify-envelope --block / eidolons run --verify) and recorded a verify_pass before dispatch. If no verify_pass exists for the message_id, or a verify_fail is present, REFUSE to process the payload and hand back to the orchestrator. Symmetric receiver gate — every Eidolon enforces it identically.

★ 2
Rynaro
Rynaro/magicite

vivi-parallel-tracks

Load ONLY under TRANCE authorization when the Plan phase yields N independent, non-overlapping implementation tracks (disjoint file sets). Operationalizes the TRANCE G4 form — parallel multi-track implementation in mandatory git-worktree isolation, max 5 clean-context tracks, a per-track verifier cascade, a non-fungible per-track ≤3 reflection budget, explicit stop conditions, and a single-threaded merge/aggregation step. Absent TRANCE gating, Vivi runs exactly as today: single-track A→P→I→V→Δ/R.

★ 2
Rynaro
Rynaro/magicite

vivi-methodology

Full Vivi cycle reference — A(nalyze) → P(lan) → I(mplement) → V(erify) → Δ(Delta)/R(eflect). Use when the task is a non-trivial feature implementation in a brownfield codebase and you need the complete methodology reference: complexity routing, evidence-grounded planning rules, test-anchoring requirements, and failure-escalation thresholds.

★ 2