| name | editorial-pipeline |
| description | Use when the user wants to run a multi-stage sequential editorial refinement on a piece of writing — typo cleanup → UK English standardisation → flow & polish → headings → proofreading/fact-checking → source/citation insertion → SEO polish. Each stage is a focused agent prompt; output of stage N is input of stage N+1. Triggers include "polish this draft", "run the editing pipeline", "writing squad", "iterative editing", "multi-pass refinement". |
Editorial Pipeline
Apply a sequence of focused editorial refinements to a draft, passing output from each stage as input to the next. The pipeline reads 7 agent prompts from ./agents/ and applies cross-cutting constraints from ./rules/ to preserve author voice, maintain detail level, and track changes.
When to use
- User asks to "polish this draft" or run iterative multi-pass editing
- Draft needs systematic refinement (typos → UK English → flow → headings → fact-checking → sources → SEO)
- User wants to inspect intermediate outputs and decide which stages to run or reorder
- Prose requires cross-cutting consistency (voice, detail, structure)
Inputs to gather
- Draft text — the source document to refine (paste, file, or workspace reference)
- Workspace path — which content-writing workspace to write intermediates to (user must have provisioned via
/content-writing:new-workspace)
- Draft name — short slug for the working subdirectory (e.g.,
blog-post-v1, opinion-piece-draft)
- Stages to run (optional) — default is all 7 in order; user may skip, reorder, or stop early
- Author voice / style notes (optional) — guidance from
rules/author-voice.md to preserve tone and register
Procedure
-
Read the rules. Load and parse rules/author-voice.md, rules/detail.md, rules/folder-structure.md, and rules/version-control.md. These constraints apply to every stage.
-
Set up working directory. Create <workspace>/editorial-pipeline/<draft-name>/ under the user's workspace data directory:
${CLAUDE_USER_DATA:-${XDG_DATA_HOME:-$HOME/.local/share}/claude-plugins}/content-writing/<workspace>/editorial-pipeline/<draft-name>/
Write the original draft to 00_original.md in this directory.
-
Run each stage in order (or user-specified order):
- Read
agents/basic-typo-reviewer.md (stage 1)
- Read
agents/uk-english-standardiser.md (stage 2)
- Read
agents/flow-and-polish.md (stage 3)
- Read
agents/headings.md (stage 4)
- Read
agents/proofreader.md (stage 5)
- Read
agents/add-sources.md (stage 6)
- Read
agents/seo-reviewer.md (stage 7)
-
Apply each agent inline. For each stage:
- Take the output of the previous stage (or the original draft for stage 1)
- Apply the agent prompt directly in this conversation
- Save the refined version to
<stage-number>_<stage-name>.md (e.g., 01_basic-typo-reviewer.md, 02_uk-english-standardiser.md)
- Enforce rules from
./rules/ — especially author voice, detail preservation, and version-control comments if the rules specify them
-
Present progression to user. After each stage, summarize changes made and offer the user options: continue to next stage, skip the next stage, reorder, or stop.
-
Final output. Save the fully refined version to final.md in the same directory. Emit a summary of all changes across stages and the location of the working directory so the user can inspect any intermediate.
Output / side effects
- Numbered intermediate files (
01_..., 02_..., etc.) written to the workspace editorial-pipeline subdirectory
final.md with the fully refined draft- User receives a summary of changes per stage and location of working directory
- User can re-run individual stages or request manual edits between stages
- No changes to the original workspace structure; intermediates are scoped to
editorial-pipeline/<draft-name>/
Safety / constraints
- Do not write outside the user's provisioned workspace. Always use the data-storage resolver.
- Do not overwrite or modify the source draft. Always work from intermediate outputs.
- Read rules (author voice, detail, folder structure) before starting; they are binding across all stages.
- User retains full control: may skip stages, pause for manual review, or request re-runs of specific stages.
- If a stage prompt references external sources (fact-checking, SEO research), note what external input is needed and ask before proceeding.
