Reasoning

Reasoning

This is a meta-skill: a rule that applies across every stage rather than belonging to one. Every domain competency in this plugin describes a domain of work; this one describes how to think while producing work in any of them — decomposing ambiguity, labeling confidence, self-critiquing before handoff, and capturing lessons. The content below is read live from this plugin's own skills/reasoning/reasoning.md every time this skill runs, so there is no separate paraphrased copy anywhere for it to drift against.

Reasoning competency (live)

!cat "${CLAUDE_PLUGIN_ROOT}/skills/reasoning/reasoning.md"

Instructions

1. Before acting on any ambiguous or underspecified input, sort what you know into fact / inference / recommendation / open question per the Decomposing An Ambiguous Request section above — do not skip straight to a conclusion. State facts plainly, label inferences as inferences (naming the facts they're built from), offer recommendations as recommendations rather than decisions already made, and never let an open question silently disappear — either resolve it and record how, or surface it explicitly.
2. Before firing ask_user_required_question for a stage's blocking_unknown_action or on_material_unknown, use the Questions To Challenge above to check whether the unknown genuinely matches a named materiality trigger, rather than asking for a merely-inconvenient one or silently skipping a genuinely material one.
3. When producing a handoff artifact someone downstream will build on without re-checking it themselves, attach a High/Medium/Low confidence label to every non-trivial claim per the Confidence Convention above, and name the specific evidence or gap behind the label in the same sentence or the next one. A claim with no label is implicitly High confidence — do not omit a label on anything you have not actually verified directly.
4. Before handing any artifact to the next stage or role, run the five-question Self-Critique Pass above and leave the answers somewhere visible (the artifact's open-questions section, a decision log entry, or the handoff message) — a self-critique pass that leaves no trace is indistinguishable, to the next reader, from one that was never run. This is separate from the role-handoff review that checks the artifact against requirements; this pass checks the reasoning that produced it.
5. When a mistake surfaces — in review, in a running system, or self-noticed — apply the Capturing a Lesson steps above: name the specific guidance that would have prevented it, decide the narrowest document type it belongs in, make the edit or record explicitly why none is needed, then confirm the edit actually closes the gap by rereading it against the original mistake.
6. If you are re-reading this skill after skills/reasoning/reasoning.md changed, trust what you just read above over any memory of a previous run — that live content is always the current source of truth for reasoning discipline in this workspace.