When the task matches no playbook, design one. The escape hatch from pstack-poteto-mode’s twenty-three bundled playbooks, and the framework’s nearest thing to a planning capability — notable because pstack otherwise refuses to plan on principle (“i don’t believe in planning. the best spec is code”). The deliverable before any code is the workflow itself.
Its routing is precise. A large or cross-cutting effort — a migration across many call sites, an ambitious multi-part change, or work the user steps away from to trust later — comes here even when a narrower playbook like Feature fits. A standing project-scale program routes to the Orchestrate playbook instead. figure-it-out designs one bespoke run; orchestrate runs the program.
- A. Frame. Do not start until the definition of done is stated as a falsifiable predicate, scope is quantified in rough units and effort with the blockers grounding surfaced, and a rigor level is chosen — biased high, with one-way doors and high blast radius getting more and reversible low-stakes steps getting less. Rigor is gates and artifacts, not “try harder”. A multi-hour run earns one checkpoint even though reversible work otherwise proceeds. This is the wiki’s most explicit pattern-scale-adaptive-planning statement: where bmad and nano-spec size the ceremony to the change, figure-it-out sizes the evidence burden to the blast radius.
- B. Design the workflow. Decompose into atomic, independently-landable units, sequenced riskiest-unknown-first. Build the verification harness before the work, with the baseline captured from the pre-change state so every check reads as “old value vs new value”. Run pstack-architect for one-way-door design decisions and skip it for mechanical work — a second arena over a settled design is over-engineering. Parallelize only across seams, each worker in its own worktree or branch. Write the designed phase list down; that list is what the human reviews.
- C. Run the loop. Each unit is an experiment: state the hypothesis, make the smallest change, measure against the predicate on the real artifact, keep it if it advanced and revert it if it did not. When something passes too easily, suspect the observation method before the system. Delegated work is paired with a judge and audited directly. A verdict is VERIFIED, NOT VERIFIED, or INCONCLUSIVE. Inconclusive is not a pass. Do not hide a negative. If a worker games the gate, reset and harden the contract; if the gate itself is wrong, fix the gate in its own change rather than routing around it.
- D. Keep the audit trail via pstack-show-me-your-work, woven through as each step lands rather than saved for the end.
- E. Verify and hand back against the Phase A predicate on the real product, not just the harness, and encode any recurring correction as a gate, lint rule, check, or script.
See Also
- pstack-show-me-your-work — the Phase D trail.
- pstack-architect — the Phase B design decision path.
- pstack-poteto-mode — routes here whenever no bundled playbook fits.
- mp-wayfinder — the counterpart that maps an oversized effort as investigation tickets; gsd-plan-phase — the plan-then-verify loop.
- stage-plan — the canonical stage this implements.