Content review
Review the given draft (or the piece the conversation is about) against the
repo's own standards. You are the check that the strategy actually made it
into the words.
Needs: nothing wired; strategy/ (messaging, personas, ICP) and
brand/voice.md filled. With a context-layer integration wired (the
Wired table in integrations/README.md), read the strategy documents it
serves and treat the Markdown files as the fallback.
Before the checks
If strategy/messaging.md, strategy/personas.md, strategy/icp.md or
brand/voice.md still carries Template: unfilled, stop: there is
nothing to review against, and a review against placeholders is confident
nonsense. Say so, suggest /setup, and offer only check 5 (mechanics)
in the meantime.
Checks, in order of severity
- Strategy fit: does the piece advance a pillar from
strategy/messaging.md, speak to a persona from
strategy/personas.md inside the fit defined by strategy/icp.md, and
stay inside the positioning? A well-written piece for the wrong audience
fails review. If a strategy file says source: context-layer, check
against the document served over MCP, not the file; if its
last_reviewed is older than 90 days, say so in the review.
- Argument: does it deliver the brief's stated argument? Flag drift
between
brief.md and the draft.
- Claims: every number and factual claim traceable (to
data/
snapshots, cited sources, or the brief's raw material). Untraceable
claims are findings, not style notes.
- Voice and terminology: against
brand/voice.md, its do/don't
examples and tone-by-context, plus the terms: product and feature names
spelled as strategy/positioning.md and brand/voice.md spell them,
the glossary and jargon the audience actually uses, and nothing from
the banned list. Quote the offending sentence and show the on-voice
rewrite.
- Mechanics: frontmatter complete and correct (
project, channel,
owner), links resolve, naming conventions per data/ontology/naming.md
if UTMs appear.
Output
Findings ordered by severity, each with location, problem, and suggested
fix. End with a verdict: ready for human review / needs work first.
Never flip status: yourself past in-review, and don't rewrite the draft
unless asked; review and revision are separate requests. The findings go
back to the author in the reply; this skill files no tasks.
1---2name: review3description: Pre-publish review of a draft against strategy, messaging and brand voice; findings only, no rewrite unless asked. Use when asked to review, check or QA a draft, or before it moves to in-review.4license: MIT5---67# Content review89Review the given draft (or the piece the conversation is about) against the10repo's own standards. You are the check that the strategy actually made it11into the words.1213Needs: nothing wired; `strategy/` (messaging, personas, ICP) and14`brand/voice.md` filled. With a `context-layer` integration wired (the15Wired table in `integrations/README.md`), read the strategy documents it16serves and treat the Markdown files as the fallback.1718## Before the checks1920If `strategy/messaging.md`, `strategy/personas.md`, `strategy/icp.md` or21`brand/voice.md` still carries `Template: unfilled`, stop: there is22nothing to review against, and a review against placeholders is confident23nonsense. Say so, suggest `/setup`, and offer only check 5 (mechanics)24in the meantime.2526## Checks, in order of severity27281. **Strategy fit**: does the piece advance a pillar from29 `strategy/messaging.md`, speak to a persona from30 `strategy/personas.md` inside the fit defined by `strategy/icp.md`, and31 stay inside the positioning? A well-written piece for the wrong audience32 fails review. If a strategy file says `source: context-layer`, check33 against the document served over MCP, not the file; if its34 `last_reviewed` is older than 90 days, say so in the review.352. **Argument**: does it deliver the brief's stated argument? Flag drift36 between `brief.md` and the draft.373. **Claims**: every number and factual claim traceable (to `data/`38 snapshots, cited sources, or the brief's raw material). Untraceable39 claims are findings, not style notes.404. **Voice and terminology**: against `brand/voice.md`, its do/don't41 examples and tone-by-context, plus the terms: product and feature names42 spelled as `strategy/positioning.md` and `brand/voice.md` spell them,43 the glossary and jargon the audience actually uses, and nothing from44 the banned list. Quote the offending sentence and show the on-voice45 rewrite.465. **Mechanics**: frontmatter complete and correct (`project`, `channel`,47 `owner`), links resolve, naming conventions per `data/ontology/naming.md`48 if UTMs appear.4950## Output5152Findings ordered by severity, each with location, problem, and suggested53fix. End with a verdict: *ready for human review* / *needs work first*.54Never flip `status:` yourself past `in-review`, and don't rewrite the draft55unless asked; review and revision are separate requests. The findings go56back to the author in the reply; this skill files no tasks.
calven-ai