critique

name	critique
description	Oppositional review of an outline or draft by an independent agent; loads piece-type rubric and audience context from project config

Critique

Stress-test an outline or draft before it moves forward. The critic is independent: it sees only the artifact, not the conversation that produced it. That independence is deliberate; the critic should not know why the author made the choices they did, only what they chose.

When to use

• The user has an outline at <drafts_dir>/<slug>/outline.md and wants oppositional review before drafting.
• The user has a draft at <posts_dir>/<slug>/index.md and wants it reviewed against its outline before revising.

Do NOT use for:

• Producing a new outline from a topic → use /interrogate.
• Writing the draft itself → use /author.
• Line-level edits on a draft without adversarial review → use /revise.

Preconditions

Run these as a silent preflight. Do not narrate them to the user unless a check fails; a passing preflight should be invisible.

1. .prospero/config.toml must exist

If .prospero/config.toml does NOT exist in the current working directory, invoke the prospero:init skill and follow it to completion. When init returns, resume at step 2 below. Do not proceed to critique with an unconfigured project.

2. .prospero/voice.md and .prospero/audience.md must be filled in

For each of .prospero/voice.md and .prospero/audience.md, compute byte-for-byte equality against the plugin's corresponding template at <plugin-root>/templates/voice.md and <plugin-root>/templates/audience.md. Use the plugin-root resolution procedure defined in the init skill's "Resolving the plugin's templates directory" section. Do not abbreviate its failure handling: zero matches → ask the user; multiple matches → ask the user; match found but voice.md or audience.md missing → halt with the resolved path.

• If either user file is Missing → halt with: ".prospero/<file> is missing. Run /init to restore it, then fill it in before /critique."
• If either user file is byte-for-byte equal to its template → halt with: ".prospero/<file> is the unmodified scaffold. Fill it in before running /critique; this is the one thing Prospero cannot write for you."
• Otherwise → the file is user content. Proceed.

3. Resolve the drafts directory and post path pattern

Load .prospero/config.toml to get the preset name. Read the preset file at <plugin-root>/presets/<preset>.toml and extract both drafts_dir and post_path_pattern. Overlay any explicit drafts_dir or post_path_pattern keys set directly in .prospero/config.toml. If no preset is named, default to plain. The resolved drafts_dir (typically drafts or _drafts) is where the outline and research file live; post_path_pattern (e.g., content/post/{slug}/index.md) is where the draft lives once /author has written it.

If config.toml is unparseable, the named preset file is missing, or the resolved preset lacks either field, halt with a message naming the specific failure and suggesting /init to reconfigure. Do not silently fall back.

Resolve the slug

Critique consumes an existing artifact, so the slug must come from the user or the filesystem:

• If the user named a file or slug directly, use it.
• Otherwise, list subdirectories of <drafts_dir>/. If exactly one exists, confirm with the author before proceeding. If multiple exist, ask which post to critique.
• If <drafts_dir>/ is empty or does not exist, halt with: "No drafts found in <drafts_dir>/. Run /interrogate first."

Mode determination

Decide which mode to run based on the artifact the author has on hand:

• Outline mode — the outline exists at <drafts_dir>/<slug>/outline.md and no draft has been written yet. This is the common post-interrogation path.
• Draft mode — the draft exists at the resolved post_path_pattern (with {slug} filled in). This is the post-author review path.

If the user named a specific file, use that file to disambiguate. If both artifacts exist and the user did not specify, ask which one to review. Do not review both in a single invocation.

If neither artifact exists for the resolved slug, halt with: "No outline or draft found at <drafts_dir>/<slug>/outline.md or the resolved post path. Run /interrogate to produce an outline first."

Process — outline mode

First, determine the piece type. If the outline explicitly names one, use that; otherwise ask the user whether the piece is an argued essay, opinion / polemic, or explainer — the three rubrics bundled with the plugin. (The bundled outline template does not record the piece type, so asking will be the common path.)

Load the matching rubric from <plugin-root>/templates/types/<type>.md, where <type> is one of argued-essay, opinion-polemic, explainer. User overrides may live at .prospero/types/<type>.md; if such a file exists, prefer it over the bundled plugin template.

Then spawn a separate agent using the Agent tool. Pass it ONLY:

• The outline path: <drafts_dir>/<slug>/outline.md.
• The rubric path you resolved above.
• The audience file path: .prospero/audience.md.
• The research file path if it exists: <drafts_dir>/<slug>/research.md.

Do NOT pass the conversation history. Give the agent this prompt body (substitute the paths):

You are an oppositional critic reviewing an outline. Your job is to find weaknesses, not to be helpful. Be direct and specific, like a code review.

First: Check today's date so you know what's current. Your training data may be stale; verify before assuming.

Read the following files in order:

1. The outline at {outline_path}.
2. The rubric at {rubric_path} — evaluate against the criteria it defines.
3. The audience context at .prospero/audience.md. The target reader is defined there; do not flag missing explanations for concepts that audience already knows. The Research Sources section lists where to verify claims and survey prior art.
4. If a research file exists at {research_path}, read it first so you don't duplicate work.

Read {research_path} first (if it exists) before running any web searches. Prior phases — interrogation, earlier critique passes — have already persisted research there; your first job is to assess whether it is sufficient for the outline's claims, not to duplicate it. Only run new web searches for specific gaps you have identified after reading what is on file. When you do search, use the sources listed in .prospero/audience.md's Research Sources section first, then broader web; do not invent sources not listed there. If the audience file has no Research Sources section, fall back to broader web search and note the absence at the top of your critique. If the existing research already covers the ground for a given claim, say so and move on.

Output a structured critique following the rubric's criteria. Be specific: "Section 3 claims X but provides no evidence" not "consider strengthening Section 3." End with a summary judgment: ready to draft, needs revision, or needs rethinking.

Append any new research findings (sources found, key quotes, URLs) to {research_path} under a section header like ## Critique session YYYY-MM-DDTHH:MM (local time, to the minute, so multiple same-day passes do not collide). Do not overwrite existing content. Create the file if missing. If you did not do any new research — everything you needed was already on file — say so under the header rather than leaving it empty.

Process — draft mode

No rubric file is loaded in draft mode; the criteria below are the rubric. Spawn a separate agent using the Agent tool. Pass it ONLY:

• The draft path, resolved from post_path_pattern with {slug} filled in.
• The outline path: <drafts_dir>/<slug>/outline.md.
• The audience file path: .prospero/audience.md.
• The research file path if it exists: <drafts_dir>/<slug>/research.md.

Do NOT pass the conversation history. Give the agent this prompt body:

You are an oppositional critic reviewing a draft against its outline. Your job is to find where the execution failed the argument.

First: Check today's date so you know what's current. Your training data may be stale; verify before assuming.

Read the following files in order:

1. The outline at {outline_path}.
2. The draft at {draft_path}.
3. The audience context at .prospero/audience.md. The target reader is defined there; do not flag missing explanations for concepts that audience already knows. The Research Sources section lists where to verify claims.
4. If a research file exists at {research_path}, read it to see what sources have already been found and check whether the draft actually uses them.

Read {research_path} first (if it exists) before running any web searches. Interrogation, earlier critique passes, and authoring have already persisted research there; your first job is to evaluate whether the draft's claims are adequately supported by research already on file, not to re-derive that work. Only run new web searches for specific gaps: an unsupported claim no source in research.md addresses, or a cited source you need to verify is real. When you do search, use the sources listed in .prospero/audience.md's Research Sources section first, then broader web. If the audience file has no Research Sources section, fall back to broader web search and note the absence at the top of your critique.

Evaluate on these five criteria:

1. Promise delivery: Does the post deliver on what the opening promises? Does the reader get what they were told they would get?
2. Evidence gaps: Where does the post assert without substantiating? Flag every claim that lacks immediate support.
3. Entry point: Does the opening earn the reader's attention in the first two sentences? Would a busy, skeptical reader keep going?
4. Section discipline: Does every section advance the thesis? Flag any section that is tangential, redundant, or could be cut without loss.
5. Coherence: Does the argument build, or does it meander? Is the logical thread clear from section to section?

For each criterion, give a rating (strong / adequate / weak) and a specific note. End with a summary judgment and a prioritized list of revisions.

Append any new research findings to {research_path} under a section header like ## Critique session YYYY-MM-DDTHH:MM (local time, to the minute, so multiple same-day passes do not collide). Do not overwrite existing content. Create the file if missing. If the draft's claims were already covered by research on file and you did no new searches, say so under the header rather than leaving it empty.

Presenting findings

Relay the agent's critique to the author verbatim. The author then chooses one of:

• Revise the artifact and re-run /critique (common when the critic flags real gaps).
• Dismiss specific criticisms and proceed (the author has context the critic lacks; that is legitimate).
• Go back to /interrogate for further development (when the outline's foundations are the problem, not its execution).

Do not argue with the author's decision. The critic is advisory.

Handoff

• After outline-mode review, when the author is satisfied, invoke /author to draft the post.
• After draft-mode review, return findings to the author for revision. Do NOT auto-invoke /revise; the author decides whether to revise, dismiss, or re-critique after their own edits.

Multiple draft-mode critique passes are expected and encouraged. A draft typically goes through several rounds of /critique → author edits → /critique before it's ready to publish; each pass tends to surface different issues as the obvious problems get resolved. Treat a second or third /critique invocation on the same draft as the normal case, not an exception. Each pass appends a new dated section to <drafts_dir>/<slug>/research.md rather than overwriting — the history is the point.