ape-poke-holes
SolidAdversarial review of design docs, specs, plans, blog posts, and technical writeups. Finds failure modes, unstated assumptions, scale cliffs, and gaps in understanding -- the questions a reader is left holding, the steps that were hand-waved, the claims that were never earned. No praise, no solutions, only holes. Trigger on "ape poke holes", "poke holes in this", "what could go wrong", "what is missing here", "red-team this", or any request for adversarial review of a design or an explanation.
Install
Quality Score: 83/100
Skill Content
Details
- Author
- arpitbbhayani
- Repository
- arpitbbhayani/ape-skills
- Created
- 3 months ago
- Last Updated
- today
- Language
- CSS
- License
- MIT
Similar Skills
Semantically similar based on skill content — not just same category
ape-review-blog
Reviews engineering blog drafts. Use this skill whenever the user says "ape review blog", "review blog", "review this post", "give feedback on draft", "what is wrong with this post", or pastes a blog draft asking for critique, feedback, or editorial review. Also trigger when the user shares a draft and asks if it is ready to publish. Always use this skill for any engineering blog review request, even if the user just says "thoughts?" after pasting a post.
adversarial-review
Adversarial critique — devil's advocate, stress-test, honest teardown ('poke holes', 'be brutal', 'was hältst du davon'); explicit request only. Routine code or design review → code-review.
adversarial-review
Adversarial review of any artifact — a plan, PRD, ADR, design doc, code diff, or the current conversation itself — from a hostile stance that tries to break it rather than improve it. Use when asked to "adversarially review", "red team this", "poke holes in", "attack this plan", "review our discussion", "what did we get wrong", "stress-test this", "steelman then break", "review this ADR/PRD/doc", or "find the flaws". Resolves the target, picks critique lenses per artifact type, self-refutes each candidate finding, and reports ranked findings with concrete failure scenarios. Read-only — never edits the artifact. Escalates to independent subagents (optionally on a different model) on request or for large targets.