Agent Skills: Document Review

Document Review

Review a requirements or plan document with a team of reviewer personas. Dispatch generic subagents, each seeded with a skill-local reviewer prompt. Synthesis sorts the findings: apply and report the ones it routes to Apply, in the document's native format, and route the rest to the user.

Done when: every dispatched reviewer returned or was named as failed in Coverage, the fixes routed to Apply are applied and reported, and the rest went through the four-option interaction (interactive) or came back as structured text with classifications intact (non-interactive).

Setup

Run this once before any subagent dispatch, and follow the directives it prints. Where one of them conflicts with this skill's own rules about asking the user questions, this skill's rules win — that holds whether the rule is scoped to a mode or global — and no blocking question is asked.

Run the fence exactly as written, as its own command: no piping, filtering, truncating, or batching. The output opens with === skill context and ends with CE_CONTEXT_END. One of those lines without the other means the output was truncated, so rerun the fence verbatim once. That is the only rerun inside this invocation; a later invocation of any skill runs its own.

SKILL_DIR="<absolute path of the directory containing the SKILL.md you just read>";
NODE="$(for c in node nodejs; do command -v "$c" >/dev/null 2>&1 && "$c" -e '' >/dev/null 2>&1 && { echo "$c"; break; }; done)";
if [ -n "$NODE" ]; then
"$NODE" "$SKILL_DIR/scripts/context.mjs" || echo "context script failed; continue with the skill's normal behavior";
else
echo "no Node runtime; continue with the skill's normal behavior";
fi

Interactive mode rules

Read references/modes.md before anything else. It owns mode detection, the non-interactive argument contract, and the question-tool rules: pre-load the host's blocking question tool at the top of the interactive flow, and fall back to a numbered list only when the harness genuinely lacks one.

Either way, a question that calls for a user decision fires the tool or falls back loudly. Narrating it as text is a bug.

Artifact Root

Resolve <root> only in the no-path interactive branch, which discovers the most recent plan under <root>/plans/. Every other run reads the document at the path it was handed. So an absolute-path or non-interactive review — /tmp/plan.md, possibly outside any repo — never depends on a repo root or a CE config it does not need.

Resolve the CE artifact root <root> before composing any artifact path.

• Read docs_root from <repo-root>/.compound-engineering/config.yaml only (<repo-root> = git rev-parse --show-toplevel). Do not read it from config.local.yaml. Unset -> <root> is docs, exactly as before.
• Validate a set value: a repo-relative directory whose real, symlink-resolved path stays inside the repo and is neither the repo root nor under .git/. Otherwise stop with an error naming docs_root and the value -- never fall back to docs.
• Use <root> as the sole artifact location: create it if absent, compose each path as <root>/<subdir> with this skill's own subdirectory, and never also read docs.

Phase 1: Get and Analyze Document

Read references/document-intake.md now. It covers how the document is obtained in each mode, the missing-document gate and its failure text, and the classification signals.

Two of its rules bound every later step.

Verify before any dispatch. Every resolved path must be readable on disk. If one is not, dispatch no personas: reviewers read from the filesystem, so they cannot reach a path that exists only on an unchecked-out branch (issue #925).

Classify by content shape and metadata, not by file path. artifact_readiness: requirements-only is a unified-requirements review — Product Contract only. A missing Planning Contract, Implementation Unit, Verification Contract, or Definition of Done is expected there, never a finding. artifact_readiness: implementation-ready is a unified-plan. Anything else takes the legacy requirements / plan split.

HTML unified artifacts take the same routes. Every fix lands in the document's native format; never insert markdown into HTML. That reference covers ID-bearing items. Pass the classification to each persona in the {document_type} slot.

Phase 2: Announce and Dispatch Personas

Read references/persona-selection.md for each conditional persona's activation signals and the announcement format. Two of those signals over-activate on plausible evidence: the sensitive-data bound on security-lens-reviewer, and the challenge-surface bar on adversarial-document-reviewer. Then read references/dispatch.md for payload variables, slicing, model tiering, and reviewer-failure handling.

The team is coherence-reviewer and feasibility-reviewer always, plus each activated conditional persona. Announce the team with a per-persona justification before any dispatch.

Dispatch generic subagents with bounded parallelism through the platform's subagent primitive. Seed each one with the full content of its references/personas/<reviewer-name>.md. Never dispatch a standalone agent by type or name.

A capacity rejection is backpressure, not reviewer failure. That reviewer stays queued and retries when a slot frees, and no reviewer is dropped because the harness cap is below the team size.

Cross-Model Judgment Pass

Run this pass if any of the conditional judgment trio was activated: adversarial-document-reviewer, product-lens-reviewer, security-lens-reviewer. Follow references/cross-model-review.md, which owns the pass end to end: host attestation, the one target and route used for the whole document, the disclosure before any egress, and how peers are launched, reaped, and folded in.

The pass is additive and non-blocking: a failure or timeout stops nothing and is named in Coverage. The checkout egress policy (cross_model_review_mode) is evaluated first and can skip the pass with a named reason. Filter recipients only when CROSS_MODEL_PEERS is set — unset means unfiltered, not unsanctioned. Never silently change an explicit model or recipient.

Phases 3-5: Synthesis, Presentation, and Next Action

Wait until every dispatched agent has returned, including any cross-model <reviewer-name>-<provider>.json returns. Then read references/synthesis-and-presentation.md. It owns the synthesis pipeline, the routing of each finding by confidence and fix class, fix application, the non-interactive envelope, and the handoff to the routing question. When promoting agreement, only an artifact with independence_verified: true counts as an independent reviewer.

Interactive mode only. Read references/walkthrough.md for the grouped confirmation, the routing question, and the per-finding walk-through. Read references/bulk-preview.md for the bulk-action preview behind best-judgment routing, Append-to-Open-Questions, and auto-resolve. Load neither before dispatch completes, and a non-interactive run never loads them at all — it stops at the synthesis envelope.

Read only the persona prompts the current review selected. The template and schema the dispatch payload fills:

@./references/subagent-template.md

@./references/findings-schema.json