agentleFS
ryanmacleanryanmaclean/skills/henshu/SKILL.md

henshu

Skill1 starsChanged 50 days agoMITView on GitHub

Use when a body of agent-generated work (specs, plans, code, translations, audit trails) has accumulated across multiple sessions and needs editorial curation — cross-checking, consistency, narrative distillation, and gap detection. Triggers include: spec sections that predate discoveries, translations not reviewed for naturalness, spool/audit trails that are audit logs but not readable stories, README that no longer matches project state, stale references after infrastructure changes.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 6 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run all five; they compound.

    line 30
  2. Distinctive:

    Read it as a native speaker would

    line 52
  3. Distinctive:

    Verify technical term consistency (pick one translation per term and use it throughout)

    line 54
  4. Distinctive:

    Check that code blocks, paths, and commands are untouched

    line 55
  5. Distinctive:

    Run: does it accurately describe the current state?

    line 64
  6. Distinctive:

    Silent deletion: Never delete an "orphaned" file without documenting why in the commit message.

    line 99

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Henshu (編集) — Editorial Curation for Agent-Generated Work

Overview

Henshu is the editorial pass that turns accumulated agent output into a coherent whole. Named for the Japanese concept of editing-as-compilation: shaping many pieces into something greater than their sum. It is distinct from code review (which judges correctness) and proofreading (which fixes surface errors) — henshu curates.

Core principle: Agent teams produce correct-in-isolation artifacts that drift from each other over time. Henshu finds and closes those gaps.

When to Use

  • Multiple agents have written to the same spec/plan across sessions
  • Infrastructure changed mid-project (hostnames, ports, paths) and docs partially updated
  • Translations were generated quickly and haven't been reviewed for naturalness
  • An audit trail (spool, changelog) exists but can't be read as a story
  • README or front-facing docs were written early and the project has moved
  • You are approaching a demo, release, or handoff and the work needs to be presentable

NOT for:

  • Real-time review of a single agent's output (use code-review skill)
  • Spelling/grammar only (overkill — just fix inline)
  • Active development where the work is still changing rapidly

The Five Henshu Tasks

Run all five; they compound.

1. Reality Reconciliation

Cross-check every claim in specs/plans against the actual codebase and infrastructure state. Flag anything that was true when written but is no longer true.

Pattern: grep for proper nouns (hostnames, ports, paths, tool names), check each against ground truth.

grep -rn "gitea.local\|:3000\|fb-vm-24\|2225\|include MINIMAL" docs/ plans/ sys/

2. Narrative Distillation

An audit trail is not a document. Distill the spool/changelog into a docs/PROJECT-LOG.md that answers:

  • What was the original goal?
  • What was tried and what was learned?
  • What decisions were made and why?
  • What is the current state?

Target: someone cold-starting in 5 minutes can understand the project arc.

3. i18n Editorial Pass

Machine-speed translations are structurally correct but often unnatural. For each translated file:

  • Read it as a native speaker would
  • Fix calque constructions (word-for-word translations that don't idiomatically land)
  • Verify technical term consistency (pick one translation per term and use it throughout)
  • Check that code blocks, paths, and commands are untouched

4. Front Door Polish

The README is the first thing a conference attendee or cold agent sees. It should:

  • Lead with the most interesting idea (not "this is a minimal FreeBSD image")
  • Show the architecture in 3-5 bullet points
  • Have working quick-start commands
  • Link to the key documents

Run: does it accurately describe the current state? If not, rewrite it.

5. Orphan Detection

Identify artifacts that no longer connect to anything:

  • Plans that describe work that was completed or abandoned
  • Memory files that contradict current state
  • Config files for infrastructure that changed
  • Translation files for documents that were significantly rewritten

Flag for removal or update; don't silently delete without a commit message.

Output Format

Henshu produces three artifacts:

  1. docs/HENSHU-REPORT.md — what was found and fixed, what was flagged for human review, what was left alone and why. Structured as: FIXED / FLAGGED / DEFERRED.
  2. docs/PROJECT-LOG.md (if one doesn't exist) — narrative distillation of the audit trail.
  3. Commits — one commit per henshu task (reality reconciliation, narrative, i18n, README, orphans). Granular commits make it easy to revert a single task if the editor disagreed.

Quick Reference

TaskInputOutputTime
Reality reconciliationspec/plan files + grepinline edits + FIXED listfast
Narrative distillationspool/changelogPROJECT-LOG.mdmedium
i18n editorial*.fr.md, *.ja.mdedited translationsslow
Front door polishREADME.mdrewritten READMEfast
Orphan detectionall tracked filesFLAGGED listfast

Common Mistakes

Over-editing: Henshu does not rewrite correct content for style. If a spec section is accurate and clear, leave it. Flag tone/style issues in the report but don't change them without user direction.

Silent deletion: Never delete an "orphaned" file without documenting why in the commit message. It may be referenced somewhere you didn't check.

Translating code blocks: Technical content in any language file — commands, paths, code, TOML — must remain identical to the English source. Henshu does not touch these.

Conflating with implementation: Henshu is editorial, not development. If it finds a genuine bug (not a documentation bug), it reports it in HENSHU-REPORT.md under FLAGGED and stops — it does not fix code.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.