Review Canvas Evidence

Review Canvas Evidence

Use this skill to challenge a workflow or progress state. The goal is to find unsupported claims, missing evidence, weak gates, schema drift, and places where a human or agent could misunderstand the plan.

Review Flow

1. Validate the workflow:

agent-canvas validate <workflow.json>

2. Inspect claims:

agent-canvas claims <workflow.json>

3. Read the workflow JSON directly when claims or statuses look suspicious.
4. Identify the workflow's pattern (see Pattern → Example Reference below) and skim the matching example. Use it as the reference shape — deviations are not automatically wrong, but they are where review attention should concentrate.
5. List findings first, ordered by severity.
6. For each finding include:

• severity,
• node ID or file/path reference,
• issue,
• impact,
• recommended fix,
• whether it blocks execution.

Evidence Semantics

Use these as the bar when reviewing claims.

progress.status values — not_started, planned, in_progress, blocked, needs_review, review_pending, completed, rejected, superseded, error. Only completed carries the load of "this is done"; all others are work-in-progress signals.

Valid evidence types — command, file, diff, test, screenshot, url, human_note. Anything else is not evidence.

Independence rule — review.status is separate from progress.status. An agent claiming completed does not constitute review approval. Look for review.status: approved distinctly.

Acceptance criteria — every node with non-trivial work should have agent.acceptanceCriteria[]. Missing criteria on a high-risk or completed node is a finding.

What To Attack

• Completed nodes with no claim evidence.
• Claims marked completed without evidence.
• Approved reviews attached to non-completion claims.
• Human gates that are bypassed or auto-resolved without human_note evidence.
• Wait gates with no release condition documented in purposeInstructions or meta.
• Parallel work that actually has shared write scope.
• Loops without stop conditions (_stopMode, _stopConditions absent or empty).
• Trycatch nodes without a defined escalation path on the catch branch.
• Branch nodes with unlabeled outputs or outputs that don't rejoin a merge.
• Missing acceptance criteria for high-risk or risk_level: high nodes.
• Schema fields silently dropped or normalized in a way that changes intent.
• Docs or summaries that disagree with the JSON.

Review Tone

Be adversarial and concrete. Do not spend time praising the workflow. If there are no blocking findings, say so clearly and still list residual risks or test gaps.

Pattern → Example Reference

Use the example matching the workflow's pattern key as the comparison baseline. Anchoring against a known-good shape sharpens findings — "this loop is missing a stop rule that single-loops_monthly-dining.json has" is more actionable than "this loop is suspicious."

All examples live under workflows/examples/. Sizes are nodes / connections.

Pattern key	Example file	Size	What it demonstrates
single-full	single-full_dinner-party.json	16 / 17	Compact "one of each" — start, parallel, merge, loop, wait, end across 7 phases.
single-loops	single-loops_monthly-dining.json	19 / 21	Iteration-heavy: 3 loops + 3 wait gates over a month-long plan with split owners.
multi-islands	multi-islands_multi-platform-publishing.json	22 / 18	Four genuinely independent chains, 4 starts → 4 ends, no shared merge.
multi-ideation	multi-ideation_syllabus-redesign.json	15 / 12	Three ideation lenses on the same problem, 3 starts → 3 ends, comparison off-canvas.
multi-merge (small)	multi-merge_landing-page.json	19 / 18	Three-start collab converging at a single merge — minimum viable multi-merge.
multi-merge (large)	multi-merge_ai-future-of-work-presentation.json	64 / 83	6 starts, 5 phases, 7 parallels + 8 merges + a trycatch. Major program shape.
branch-decision	branch-decision_paper-submission.json	9 / 9	Minimal binary branch (READY vs. REVISE) with rejoin at a merge.
trycatch-escalation	trycatch-escalation_grant-application.json	10 / 9	Trycatch wrapping a funder decision — funded path activates, rejected path escalates.
subflow-composition	subflow-composition_course-launch.json	10 / 10	Four subflow markers signaling "this step is its own workflow."
pipeline (full)	greenwashing.json	66 / 91	Real working lit-review pipeline — 9 phases, parallel fan-outs, loops, human-in-the-loop wait.
pipeline (compact)	greenwashing-literature-review-multi-start.json	13 / 13	Slim multi-start lit-review variant.
cop (legacy)	cop.json	—	Original community-of-practice reference example.

Useful References

• skills/plan/SKILL.md — design contract (full pattern catalog, design recipe, gate semantics)
• SCHEMA.md — JSON schema reference
• AGENT-CANVAS.md — agent operating handbook
• USER-GUIDE.md — browser canvas guide