| name | feasibility-attack |
| description | Attack a proposal's feasibility before endorsing it: extract its binding requirements — resources, limits, permissions, dependencies, deadlines — and test each against known constraints. One hard infeasibility ends the endorsement. |
| license | MIT |
feasibility-attack
Trigger (observable): A novel proposal, plan, or idea is being evaluated or about to be endorsed, and it carries no cost, resource, or constraint check — enthusiasm and mechanism are present, arithmetic is not.
When NOT to activate: The proposal already includes a constraint-by-constraint feasibility case (then verify that case instead of rebuilding it); pure ideation where feasibility filtering was explicitly deferred; proposals already killed on other grounds.
Procedure
- Extract the proposal's binding requirements: budget, compute, headcount-time, API/rate limits, permissions, physical constraints, hard deadlines, and dependencies on parties who haven't agreed.
- For each requirement, test it against a known constraint with arithmetic or a citation — 'needs 40 req/s; the vendor caps at 10/s' — not vibes.
- Distinguish hard infeasibility (violates a limit that cannot move) from soft (expensive but purchasable). One hard hit ends the endorsement.
- For soft hits, state the price of relief (time, money, negotiation) so the proposer decides with the bill visible.
- Close constructively: the cheapest modification that restores feasibility, or a clean kill with the binding constraint named — never a defect list with no verdict.
Required output
A requirement-by-constraint table with pass / soft-hit (priced) / hard-hit (killing) results, ending in an explicit feasible / feasible-at-cost / infeasible verdict.
Verification
- Every hard kill cites a named constraint with arithmetic or a source; assumed constraints cannot carry hard kills.
- The output ends in an explicit feasible / feasible-at-cost / infeasible verdict that follows mechanically from the table.
- Net effect: every binding requirement is tested against a named constraint, and the verdict follows from the hits.
Known risk: Killing ideas on assumed constraints that were never checked ('the API probably caps at...'). Mitigation: each constraint must be cited or computed, or it's labeled assumed and can't carry a hard kill.
Max intended cost: ≤350 added output tokens; a few lookups to confirm claimed limits.
Evidence status: DESIGNED — specified from documented reasoning-failure modes; not yet executed as a packaged skill.
Lineage: Derived from a documented reasoning-failure mode — endorsing novel ideas on mechanism and enthusiasm without feasibility arithmetic — combined with two evidence-backed principles: close with an explicit verdict rather than an unranked defect list, and recognize when a constraint space is exhausted rather than continuing to search it.
