Menu
Adversarial pre-build pass that challenges a plan before execution begins. Five angles, one verdict.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Use when a long-form manuscript (book chapter, ebook, multi-chapter playbook, long-form digital product) needs an editing pass to identify and remove AI writing tells. Sweeps across 8 pattern categories, assembles a structured edit packet for operator review, and applies approved edits.
Use when you have spare capacity and want to set your system improving without active attention. Operator-invoked only, never scheduled. Safe, additive, reversible hygiene work runs under a threshold model with a permanent floor of actions the agent never takes on its own.
Use when archiving several workspace artifacts at once with per-artifact approval gates before any archive move executes.
Use when deciding whether a piece of deterministic, no-judgment work should route to a secondary AI provider instead of your primary interface. Covers the provider catalog, the five-condition eligibility test and the dispatch script.
Retire a stalled or indefinitely-deferred plan without losing its value. Candidate scan, harvest, distribute, sweep, archive and codify.
Automated pull request review for your repos. Five parallel agents, confidence scoring, convention-file compliance, and GitHub comment posting.
| name | plan-challenger |
| description | Adversarial pre-build pass that challenges a plan before execution begins. Five angles, one verdict. |
| status | active |
| version | 1.1 |
Purpose: Surface the hard questions before a plan goes into execution. Not a veto. A mirror. Five structured angles against the plan, with a one-line verdict at the end. Trigger: The operator invokes "Plan Challenger" with a plan name or plan content. Also surfaces as a one-line opt-in inside Pending Plan Implementation after the Pre-Execution Checklist. Inputs: Plan content (name to locate, or paste directly). Outputs: Five-angle challenge report, taste-decision flags for near-call questions, one-line verdict. Status: active
Read the target plan. If a name was given, locate it in your plans directory. If content was pasted, use it directly.
Confirm the plan title and intent back to the operator in one sentence before proceeding.
Run each angle in sequence. Each angle produces one to three sharp observations or questions, not a list of everything that could go wrong. Surface what matters most.
Angle 1: Strategic Fit Is this the right problem for this moment? What else is actively competing for this slot? Does this plan advance the operator's current primary arc, or does it branch sideways?
Angle 2: Timing Why now? What changes if this is deferred two weeks? What external factors or internal pressures are driving the timing? Is the urgency real or inherited from habit?
Angle 3: Baked-In Assumptions What must be true for this plan to succeed that has not been verified? Name the assumptions explicitly. Rate the risk of each: low (safe to proceed), medium (worth a quick check), high (should be resolved before execution begins).
Angle 4: Architectural Risk What is the most likely way this plan breaks or creates debt? What part of the implementation is most fragile? What does success look like on paper but create pain in six months?
Angle 5: Opportunity Cost What are we not doing by doing this? What capability, relationship or momentum is being paused or sacrificed to run this plan now?
After the five angles, scan for near-call questions. Places where two framings are genuinely close, scope is borderline, or a reasonable case exists for a different direction. Surface these explicitly rather than collapsing to one answer.
Format: "Close call: [question]. Both [framing A] and [framing B] are defensible. The operator decides."
Surface only genuine near-calls. Do not manufacture uncertainty where clarity exists.
One line. No hedging.
The verdict is a recommendation, not a gate. The operator decides.
For high-stakes plans, assign a confidence score (0-100) to each finding from the five angles. This converts the challenge from a list of objections to a weighted list where severity is explicit.
Use this rubric (pass verbatim to any sub-agent doing the scoring):
0: Not confident at all. This concern does not hold up to scrutiny or is already addressed in the plan.
25: Somewhat confident. This might be a real concern but may also resolve on closer inspection. Worth noting, not blocking.
50: Moderately confident. This is a real concern but a minor one. It would not change the execution path.
75: Highly confident. This concern is very likely real and will affect the plan in practice. The existing approach is insufficient without addressing it.
100: Absolutely certain. This concern is definitely real, confirmed by evidence in the plan, and will block or break execution if unresolved.
Filter: any finding scored below 50 is a note, not a challenge. Any finding scored 75 or above warrants a named action in the verdict.
This scoring layer is optional. Use it when the five angles produce many findings and the operator needs a ranked view before deciding. On straightforward plans, qualitative assessment is sufficient.
Dispatch the cheapest model that does the job well.
| Step | Default model | Rationale |
|---|---|---|
| Step 1: Load the plan | Haiku | File read and one-sentence intent confirmation; no judgment required |
| Step 2: Five-angle challenge | Opus | Adversarial architectural judgment across five distinct lenses; depth over breadth demands the strongest model |
| Step 3: Taste-decision surfacing | Opus | Identifying genuine near-calls requires nuanced judgment; false certainty here harms decision quality |
| Step 4: Verdict | Opus | One-line verdict is an operator-facing judgment call; precision and accountability make Opus the right tier |
| Confidence scoring (optional) | Sonnet | Applying a rubric to score existing findings is structured evaluation, not fresh architectural judgment |
Set the model explicitly on every subagent dispatch. Never silently inherit the top tier.
Source Harvest is the gateway skill for systematic pattern extraction from external repos and tools. Many users adopt Source Harvest first, then layer additional skills like this one on top.
Pending Plan Implementation executes a plan after challenge. Strong pairing for high-stakes work: challenge first, then implement.
Two-skill handoff protocol: after the Pre-Execution Checklist in Pending Plan Implementation, a one-line opt-in appears: "Challenge warranted?" If yes, invoke Plan Challenger with the plan content, work through findings, then return to Pending Plan Implementation Step 1. If no, proceed immediately.
If Source Harvest isn't installed yet: Install Source Harvest via IGOS. If Pending Plan Implementation isn't installed yet: Install Pending Plan Implementation via IGOS.
(Empty. Populated when execution mistakes occur during sessions.)
