| Field | \nBound contract | \n
|---|---|
| Trigger | \nUser wants adversarial stress-testing of an architecture, structure, or shape. Either the agent attacks and the human defends, or the agent proposes and the human attacks. | \n
| Authority | \nRead-only. No file, VCS, credential, paid, published, deployed, or remote mutation. The agent emits chat output only. | \n
| Side effect | \nAn attack or proposal transcript in chat, plus the surviving or broken shape recorded in chat. | \n
| Done | \nThe shape has been attacked from both directions, the human has defended or attacked, and the surviving shape (or non-converged record) is recorded. | \n
The proposed architecture, structure, or shape to test. The human must supply it as text, a diagram description, or a reference to an artifact already in context. Without it the skill stops.
\nOptional: named attack axes the human wants prioritized (coupling, abstraction depth, failure modes, scalability, ownership boundaries).
\nOptional: a scope limit naming what is out of bounds for the attack.
\nOptional: constraints, scale, existing components, and non-goals for a proposal round.
\nA human participant must be present; the skill does not converge without one.
\nRestate the target shape from the supplied input and confirm scope with the human before attacking or proposing. Do not test material outside the stated scope. Done when: the shape is restated and scope is confirmed.
\nDetermine the direction. If the human supplied a shape to attack, the agent attacks and the human defends. If the human asked the agent to propose, the agent proposes and the human attacks. Done when: the direction is set and the first move is made.
\nEnumerate the attack axes relevant to the shape: coupling between parts, abstraction depth versus leakage, failure and recovery paths, scalability and growth pressure, ownership and boundary clarity, plus any named axes the human supplied. Done when: all relevant axes are enumerated.
\nMount the first move. In attack mode: for each axis, state the specific structural weakness, the condition under which it fails, and the evidence. Each attack must be falsifiable by the human's defense. In propose mode: propose one architecture shape with components, boundaries, data flow, and the single load-bearing decision that makes it cohere, plus the failure mode each component rejects. Done when: each attack states weakness, failure condition, and evidence, or the proposal names components, boundaries, data flow, the load-bearing decision, and per-component rejected failure modes.
\nPresent the moves and invite the human to respond. Wait for the human's defense or attack before judging survival on any axis. Done when: the human is invited and the agent waits.
\nFor each response, classify the outcome: rebutted (the shape survives that axis), landed (the shape must change on that axis), or deferred (insufficient information to judge). Done when: each axis is classified.
\nWhen an attack lands or a human attack breaks a proposal, propose the minimal shape change that resolves the weakness and re-test the changed portion once. Done when: a minimal shape change is proposed and re-tested once.
\nRepeat steps 3 through 7 until every axis is either rebutted or resolved, or the human ends the session. Done when: every axis is rebutted or resolved, or the human ends the session.
\nRecord the surviving shape: the final structure, the attacks it survived with their rebuttals, the changes made, and any deferred or non-converged axes with their open questions. Done when: the final structure, survived attacks, changes, and deferred axes are recorded.
\nNo target shape supplied and no proposal requested. Stop and ask the human for the shape and scope. Do not invent a shape to attack or propose.
\nAttack not falsifiable on an axis. Classify that axis as deferred with the open question. Do not claim it survived.
\nHuman response absent or non-substantive for a landed attack. Mark the axis as unresolved. Do not declare the shape surviving on that axis.
\nScope widening. If the human introduces new material mid-session, restate the new scope and confirm before continuing. Never silently expand the attack surface.
\nNon-convergence. If the same axis oscillates without resolution across two re-tests, record it as non-converged with the conflicting positions rather than forcing a verdict.
\nNo mutation. This skill writes nothing to disk or VCS. Recovery is to restate the surviving or non-converged shape in chat.
\nA surviving-shape record in chat containing the final structure, the list of attacks it survived with their rebuttals, the changes made during the session, and any deferred or non-converged axes with their open questions. If the shape did not survive, output the non-converged record naming the unresolved axes.
\n","files":[{"path":"agents/openai.yaml","sizeBytes":165,"isText":true},{"path":"SKILL.md","sizeBytes":5118,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-09-30T19:50:28.998527Z","sha256":"9BA6EEE2DC1856496598E74D5B1F95F40EE2759D06E69A7A2B36CA5A3D238D67","sizeBytes":2327},"review":null,"source":{"repositoryUrl":"https://github.com/OutlineDriven/outline-driven-development","path":".devin/skills/attack-shape","license":"Apache-2.0","commit":"b0e8ce89a19fac880251dc3ea1babfeb4503a4fe","subtreeSha":"3CF01EC3EB615ED2C08640DFFEEC8444F8B85F96BD1DEFC4EFDEA8A4A623FAFF","lastSyncedAt":"2026-09-30T19:49:48.917811Z"},"reviewedAt":"2026-09-30T19:51:37.888701Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/OutlineDriven/outline-driven-development/tree/main/.devin/skills/attack-shape"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install outlinedriven-outline-driven-development@llmmart"},{"target":"git","command":"git clone https://github.com/OutlineDriven/outline-driven-development.git"}]}