--- name: report-review description: Review academic or professional reports, consolidate feedback into an iteration plan, execute an approved revision, or guide a verified manual walkthrough. Use for whole-report review workflows, not isolated sentence edits. --- # Report review Coordinate a review whose depth is visible and whose scope follows the user's request. This skill works for research, technical, business, and reflective reports; determine the genre before applying conventions. ## Recurring issues Whenever a user or reviewer identifies a complex or systematic issue that plausibly recurs, formulate or refine a contextual pattern and record the example. Apply the pattern across the draft when that broader check is already authorised; otherwise explicitly offer the wider pass alongside addressing the local request. Follow the [recurring-pattern procedure](references/recurring-patterns.md) for scope, contextual matching, counterexamples and completion. This applies even when only one instance has been pointed out; a routine isolated typo does not automatically justify a broad audit. ## Establish the review contract Identify the current editable source, canonical export, review records, intended audience, assignment or publication requirements, and applicable rubric. Verify file identity rather than choosing whichever filename says final. Preserve an appropriate baseline before substantial changes and inspect existing user edits. Distinguish review-only, planning, authorised editing, and manual walkthrough. Follow existing authorisation without repeated permission requests. During review-only work, save findings without changing the manuscript. Preserve user exclusions and deferred material. Disclosure requirements come from the actual institution or venue; surface a conflict rather than inventing a personal acknowledgement or treating a historical deferral as universal. Read [review records](references/review-records.md) when creating an itinerary, audit trail, or walkthrough queue. Reuse existing project records instead of adding competing trackers. Store document-specific decisions there; skill installation alone does not preserve a report's state across chats. ## Route substantive work For a full deep pass, inventory all sections and ancillary material, including abstract, appendices, tables, figures, listings, captions, notes, and references. Read each in context; searches discover candidates but cannot establish full coverage. Map every requested pass to its document coverage and outcome; do not silently substitute a few familiar examples for a full audit. Include headings, footnotes, labels and legends in the inventory. Use the installed companion skills when their work is in scope: - [Report clarity](../report-clarity/SKILL.md): reader assumptions, terminology, flow, relevance, internal navigation, and prose review. - [Report evidence](../report-evidence/SKILL.md): claims, provenance, comparison fairness, sources, bibliography, and external links. - [Report release](../report-release/SKILL.md): rebuilding, rendered verification, final identities, and passage-page mapping. A narrow request limits the work performed; offer a wider check when the recurring-issue rule applies. For a full authorised iteration, resolve substantive questions and structure first, then run the writing passes, then verify the rendered release. If a companion is unavailable, report the specific missing capability and use available equivalent methods without claiming that skill ran. ## Requirements and rubric depth Read the actual brief, complete rubric, template and submission instructions. Separate mandatory constraints, indicative targets, and optional style advice. Establish which authority governs each requirement for this deliverable. Record missing or conflicting governing documents and limit conformance conclusions to checks supported by the available evidence. For a standards-led review, use the [standards guide](../report-drafting/references/report-standards.md) to find official sources. Reuse the drafting requirements record if present and verify the governing editions before assessing coverage. Map every applicable criterion to a passage and supporting evidence. Assess justification, synthesis, critical analysis and completeness at the requested level, rather than marking a criterion met because its heading exists. Recheck after cuts. Report missing research or unavailable requirements separately from editorial defects; prose cannot establish an experiment, grade, or independent validation. ## Build and execute the itinerary Reconcile prior feedback into exact approved spans, approved directions, unresolved factual questions, intentional retention, and deferred items. Protect only the approved span, not every adjacent sentence. If exact approved wording contradicts evidence, explain the conflict rather than silently preserving or rewriting it. Make each task concrete: location, issue, intended result, dependency, evidence needed, and completion check. A previous iteration's pass is background, not proof a fresh pass ran. If a new full pass was requested alongside known fixes, schedule and perform both. Carry adopted conventions forward to new and revised material within this report, rather than treating them as a one-iteration checklist. Record actual findings and justified no-change decisions. After technical changes, perform the clarity skill's editorial sequence and any explicitly selected available writing skills. Recheck meaning and evidence after prose edits. Keep the report's genre and voice: a reflection needs credible personal learning and future action, not an imposed research-paper structure. Never fabricate personal experiences, decisions or feelings. ## Manual walkthrough and stopping For a manual walkthrough, read the [walkthrough procedure](references/review-records.md#walkthrough-queue). Use verified passage locations, one bounded question per stop, actual remaining counts, and the user's requested order. Deliver requested wording or queues visibly; distinguish user acceptance from agent verification. Finish when in-scope findings have supported dispositions and required verification is complete. Separate optional polish from blockers. State coverage and limitations honestly, and deliver source/export links, a concise change summary, and only genuinely outstanding decisions. No submission, publication or new experiments are implied by editorial authorisation.