From memento
Step 1 of Memento. Use when receiving a new task (markdown, screenshots, verbal description) that will go through the Memento cycle, before any planning or code.
How this skill is triggered — by the user, by Claude, or both
Slash command
/memento:memento-1-brainstormingThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Explore intent, requirements, constraints. No code yet. No plan file yet.
Explore intent, requirements, constraints. No code yet. No plan file yet.
The most expensive misses aren't bad design — they're a wrong frame inherited from the task and environment facts assumed from partial code reads. Propose-4-options refines inside a frame; it cannot break the frame. The auto-review skeptic (step 3) attacks the plan within locked decisions — by then it's too late. So before any decision sets, dispatch one Premise Auditor subagent (Explore-class), in parallel with the memento-0 prior-art sweep.
Independence is the mechanism. Give it the raw task (the user's original words) + repo access — never your already-narrowed summary. Brief it plainly: "the framing in this task may be wrong; assume nothing the task asserts is true until checked." If you feed it your frame, it inherits your blind spot. (Same reason test-writer ≠ implementer.)
Mandate — attack the premise + environment, NOT the design:
CLAUDE.md was flagged stale (step 0c), or it reads stale, include its load-bearing claims among the assumptions you check; you read code, so verify each against the current code. A claim the code contradicts is a confidence blocker, reported with its where-to-check like any other.Output budget (hard): return only —
{confidence, falsifier, where-to-check}.No 20-item doubt dumps — an unscoped skeptic trains you to ignore it.
Gate: brainstorming may not proceed to propose-4-options until the Auditor's human-only questions have been asked and the environment/topology facts it flagged are confirmed (repo evidence or human). A frame-breaking finding loops scope back, not forward.
This is a pre-decision instrument only. Do not re-run it as a standing reviewer in later steps — post-decision, it's just noise on top of auto-review.
memento-0 ran a prior-art sweep before sizing. If it found an existing/legacy implementation, scope against that — do not re-derive the feature from the surface request. If brainstorming uncovers a domain area the sweep did not cover, dispatch a targeted follow-up Explore sweep before concluding.A shared understanding of what the user wants and why. No artifact written yet — that happens in step 2 (memento-2-planning).
Premise gate. The Premise Auditor must have run, its human-only questions been asked, and the environment/topology facts it flagged been confirmed (repo evidence or human). An unresolved frame-breaking finding loops scope back — it does not transition.
Confidence gate. Before transitioning to planning, every domain or business-rule claim below HIGH confidence must be confirmed against an authoritative source — the team leader, a written spec, or the prior implementation. List the unconfirmed claims to the user and resolve them; do not start planning on top of unconfirmed business rules. A CLAUDE.md claim flagged stale (step 0c or the Auditor) resolves by additive reconciliation: edit the doc to add missing facts, fix provably-wrong references/values, and annotate contradictions, while preserving the existing human prose and structure — never wholesale-rewrite, reorder, or recreate it. The user may defer the doc fix and proceed.
When user signals "ok, plan it" (or equivalent) and the confidence gate is clear, invoke memento-2-planning.
npx claudepluginhub joaoant42/memento --plugin mementoGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.