PM Method

Elicit the real requirements by interviewing the requester BEFORE building or writing anything — one question at a time, until the brief is buildable. Use when a request is vague ('make me a dashboard', 'write something for the board'), when past deliverables missed the mark, or when the user says 'interview me' / 'ask me questions first'. Produces a validated brief: goal, audience, constraints, success criteria, and explicit non-goals — then, and only then, the work.

Run a real brainstorm — divergent generation without judgment, then convergent selection with explicit criteria — instead of listing ten obvious ideas and calling it creativity. Use when asked to brainstorm, generate ideas or options, explore a solution space, or name something. Produces a genuinely wide option set (including the weird tail), then a shortlist selected against named criteria with the rejects preserved.

Write an executable work plan BEFORE starting a complex task — decomposed steps with verification points, risks pre-named, and explicit stop conditions — so execution becomes checking boxes instead of improvising. Use when a task will take many steps, when asked to plan before doing, when previous attempts sprawled or stalled, or before delegating work to subagents. Produces a plan document another agent (or future you) could execute without re-deriving the thinking. Pairs with executing-plans.

Audit any document against its own sources — every factual claim extracted and graded as evidenced, partially evidenced, unsupported, or contradicted, with the exact source line that supports or fails it. Use when asked to fact-check a document against its sources, check whether a report's claims are backed up, verify a deck against the data, or ask 'does this doc have receipts?'. Produces a claim ledger, unsupported claims ranked by load-bearingness, a fix-or-drop call per claim, and an honesty score with stated method.

Execute a written plan with discipline — verify each step before advancing, surface deviations instead of improvising around them, and keep a visible execution log. Use when working through a plan (yours or another agent's), resuming multi-session work, or when execution keeps drifting from what was agreed. Produces completed work plus an execution log showing what matched the plan, what deviated and why, and what the plan got wrong. Pairs with writing-plans.

Reconstruct the decision actually made in a messy Slack, email, or meeting thread into a proper decision record — commitments named, silent assumptions surfaced, non-decisions called out. Use when asked what did we actually decide, turn this thread into a decision record, who committed to what, or reconstruct this discussion. Produces a decision record with quoted evidence, a commitments table, reconstructed assumptions, dismissed options, and a confidence note on the reconstruction itself.

Simplify code that works — remove speculative abstraction, dead flexibility, and needless indirection while keeping behaviour identical and verified. Use after a feature lands ('now simplify it'), when AI-generated code arrives over-engineered, when a file has grown hard to follow, or as the cleanup pass before review. Produces a smaller, flatter version with identical behaviour, plus a ledger of what was removed and why it was safe. For finding bugs use code-review-checklist / ai-code-review — this skill assumes it works and makes it simple.

Decompose work across parallel subagents properly — task slicing that avoids collisions, briefs that stand alone, and result integration that catches contradictions. Use when work can genuinely parallelise (research fan-outs, multi-file changes, independent analyses), when deciding whether to delegate or do it yourself, or when past multi-agent runs produced conflicts and duplicated effort. Produces an orchestration plan: the parallel/sequential split, per-agent briefs, and the integration protocol.

Build in small, individually-verified increments that each leave the system working — instead of big-bang changes that fail mysteriously at the end. Use when implementing multi-part features, refactoring anything load-bearing, making large mechanical changes, or when past work produced huge diffs that were wrong somewhere unfindable. Produces the same end state as the big bang, reached through verified checkpoints you can stop at, ship from, or roll back to.

Verify work actually meets its brief BEFORE declaring it done — a structured self-review pass that catches the gaps, unmet requirements, and untested claims that 'looks finished' hides. Use before handing over any deliverable (document, code, analysis, plan), when past work kept coming back with 'you missed…', or as the standing final step of any multi-step task. Produces the verified deliverable plus a short verification record: what was checked, what was found and fixed, what remains open.