prose-polish

Prose Polish

Refactor natural-language prose into concise, native-sounding text in a target language, using a sonnet subagent by default (the model id is overridable). The refactoring runs in a fresh Agent dispatch so the executor judges the prose without the main thread's context; the main thread applies the result. This is a single-pass skill.

Two modes (mutually exclusive, selected by which inputs the caller supplies — see ## Invocation contract § Mode determination):

• File mode — given one or more file paths, rewrite the target-language natural-language prose in place (comments, test descriptions, docstrings, user-facing strings). Code, identifiers, proper-noun product / API / library / tool names, and logic-bearing string literals are left untouched; ordinary technical vocabulary sitting inside the target-language prose is translated, not preserved.
• Text mode — given a block of text, return the refactored text (for polishing prose before it is presented).

Designed to be called from non-interactive routines; it never prompts the user — it either returns a structured summary or terminates early with a machine-readable reason code.

Invocation contract

The caller passes these fields in natural language (the skill extracts them from the invocation text). A field counts as provided iff the caller supplied a non-empty, non-whitespace value; empty string and whitespace-only count as absent.

• File: / Files: (file mode — one or more relative paths) — the files whose target-language prose is rewritten in place. Multiple paths may be listed (one per line or comma-separated).
• Text: (text mode — the prose to refactor) — the block of text to polish and return.
• Language: (optional, default ja, e.g. ja / en) — the target language whose prose is refactored. In file mode, only prose written in this language is rewritten; prose in other languages is left untouched.
• Model: (optional, default sonnet) — the model id applied as the model parameter on the refactor Agent dispatch (Step 3 (a)). Validity predicate: a value is valid only if it is one of the model ids the current Agent tool's model parameter accepts — check the tool's live schema loaded in the current session rather than a fixed list, since Anthropic adds new model families over time (sonnet / opus / haiku / fable as of this writing); a full claude-* id (e.g. claude-sonnet-5) is outside that parameter's accepted aliases and is therefore invalid too. An absent field or an invalid value falls back to the default sonnet — sonnet produces more concise, natural prose than the larger models this skill is meant to clean up after.

Pass related files together (file mode) — cross-file duplicate-comment detection (## Process Step 3's recommendations) works only when the related files are listed together in a single file-mode invocation: the refactor subagent can spot a comment duplicated across files only when it sees those files in one dispatch.

Mode determination

Evaluate against the two mode selectors — the File: / Files: group and Text: — using the provided/absent rule above:

• File: / Files: provided AND Text: absent → file mode (run ## Process Steps 1–4 in the file-mode branch).
• Text: provided AND File: / Files: absent → text mode (run ## Process Steps 1–4 in the text-mode branch).
• Both provided → return early with {"status": "error", "mode": null, "language": "<resolved>", "applied_edits_count": 0, "files_modified": [], "recommendations": [], "refactored_text": null, "reason": "ambiguous args"} — two modes were requested at once; surfaced loudly rather than silently picking one.
• Both absent → return early with {"status": "error", "mode": null, "language": "<resolved>", "applied_edits_count": 0, "files_modified": [], "recommendations": [], "refactored_text": null, "reason": "incomplete args"} — no mode could be selected.

This fixed mode gate (one selector group present → that mode; both → ambiguous; neither → incomplete) surfaces a conflicting or empty argument set as a loud error rather than silently picking a mode. On both early-return errors mode is null (no mode was selected); callers branch on status == "error" + reason.

Process

Step 1 — Determine mode and parse inputs (main thread)

1. Resolve Language: to <resolved-language> — the provided value, else the default ja. This resolved value is echoed in the return contract's language field (so a caller that passed nothing can tell the skill defaulted to ja).
2. Parse the optional Model: value per § Invocation contract's Model field — hold a valid value for the Step 3 (a) dispatch; absent or invalid → default sonnet.
3. Determine the mode per § Invocation contract § Mode determination. On ambiguous args / incomplete args, emit the corresponding early-return verdict and stop.
4. File mode: collect the listed paths into target_files (the scope-check baseline for Step 3 (b)). Text mode: hold the input text as input_text.

Step 2 — Load the style guide (main thread)

Read references/prose-style-guide.md — the concise-and-natural prose rules injected into the dispatch payload (Step 3 (a)). In file mode, also Read each entry in target_files for injection into that payload.

Step 3 — Dispatch the refactor subagent

(a) Dispatch

Dispatch a fresh subagent via the Agent tool (subagent_type: general-purpose), passing the parsed Model value as the Agent model parameter (the default sonnet when none was provided). Assemble the dispatch prompt from the sections below, each framed with a clear --- LABEL --- fence so the subagent can parse each payload unambiguously:

• --- PROSE STYLE GUIDE ---: the full content of references/prose-style-guide.md
• --- TARGET LANGUAGE ---: the <resolved-language> code
• File mode — --- TARGET FILES ---: each entry in target_files as a ### <path> sub-heading followed by the file's full current contents
• Text mode — --- INPUT TEXT ---: the input_text verbatim
• --- REFACTOR PROMPT ---: the mode-appropriate prompt below (verbatim)
• --- RESPONSE FORMAT ---: the mode-appropriate response format below (verbatim)

Refactor prompt — file mode (include verbatim in the dispatch):

You are a fresh prose editor. You have not seen prior conversation context — only the PROSE STYLE GUIDE, TARGET LANGUAGE, and TARGET FILES below. For each TARGET FILE, find natural-language prose written in the target language — code comments, test / example descriptions, docstrings, and user-facing string literals — and rewrite each one to be concise and natural for a native reader of that language, following the PROSE STYLE GUIDE.

Preserve everything that is not target-language prose (hard constraint): never change code, identifiers, function / variable / type names, proper-noun product / API / library / tool names and code symbols, import paths, or any string literal that carries program logic (keys, enum values, format specifiers, paths, commands). An ordinary source-language word sitting inside the target-language prose — a common verb, noun, or adjective with a natural target-language equivalent, not a proper noun or code symbol — is itself translatable prose, not a preserved token: render it in the target language per the PROSE STYLE GUIDE's Preserve-vs-translate litmus test rather than leaving it code-mixed. Leave a whole passage written entirely in another language untouched. If a candidate change could alter program behavior or touch a non-prose token, do not emit it.

Return each rewrite as a {file, old_string, new_string, rationale} Edit. old_string must match exactly one location in the current file — include 1–3 lines of surrounding context so the snippet is unique (short one-liners collide and cause the Edit to fail). A rewrite may reduce the number of prose lines: merge adjacent comment lines that state the same thing, or delete a comment whose only content is what-narration of the code beneath it (per the style guide's "say what the code does not" rule) by emitting an edit whose new_string omits that line — this is a prose change, not a structural code edit. Delete a comment only when it is fully redundant with the adjacent code; otherwise shorten it. When old_string carries a non-target line purely for uniqueness (an adjacent line in another language, or a code line), reproduce that line byte-identically in new_string so the preserve constraint is not breached. If a file needs no prose changes, emit no edits for it. If nothing needs changing across all files, return edits: [].

Cross-file duplicate comments → a recommendations entry, not per-copy edits: when a comment qualifies as a cross-file duplicate under the PROSE STYLE GUIDE's Cross-file duplicate comments rule (which owns what qualifies, the exclusions, and the threshold), do not emit a per-copy polish edit for those copies — instead emit a single recommendations entry (see RESPONSE FORMAT) flagging the duplication. A comment that does not qualify is ordinary prose — polish it as usual. If nothing qualifies, return recommendations: [].

Response format — file mode (include verbatim in the dispatch):

Write your reasoning briefly, then end your response with a single fenced JSON block matching this schema:

```json
{
  "edits": [
    {"file": "<path>", "old_string": "<unique 1-3 line snippet>", "new_string": "<replacement>", "rationale": "<short reason>"}
  ],
  "recommendations": [
    {"summary": "<one-line description of the duplicated knowledge>", "files": ["<path>", "<path>"], "suggestion": "<consolidate-into-one-place-and-remove-inline-copies advice>"}
  ]
}
```

recommendations holds cross-file duplicate-comment consolidation candidates (return [] when none qualify): summary identifies the duplicated knowledge in one line, files lists the two or more TARGET FILES the comment recurs in (the set, not ranked), and suggestion is the concrete consolidate-and-remove-copies advice — when it names a consolidation destination, phrase it as an illustrative example (e.g. a shared doc or rule file) rather than asserting a specific path, since the skill never verifies or dereferences it.

Refactor prompt — text mode (include verbatim in the dispatch):

You are a fresh prose editor. You have not seen prior conversation context — only the PROSE STYLE GUIDE, TARGET LANGUAGE, and INPUT TEXT below. Rewrite the INPUT TEXT to be concise and natural for a native reader of the target language, following the PROSE STYLE GUIDE.

Preserve non-prose tokens (hard constraint): keep identifiers, code fragments, proper-noun product / API / library / tool names and code symbols, file paths, and command / config strings verbatim — refactor the natural-language wording around them. An ordinary source-language word inside the target-language prose — a common verb / noun / adjective with a natural target-language equivalent, not a proper noun or code symbol — is translatable prose: render it in the target language per the PROSE STYLE GUIDE's Preserve-vs-translate litmus test, not left code-mixed. Do not add, drop, or reorder factual content; only improve concision and naturalness. If the text is already concise and natural, return it unchanged.

Response format — text mode (include verbatim in the dispatch):

End your response with a single fenced JSON block matching this schema, and write no other prose:

```json
{
  "refactored_text": "<the rewritten text>"
}
```

Agent-unavailable fallback: detect availability by inspecting the current tool surface — do not attempt a speculative call to probe it. When the Agent tool is absent (e.g. this skill runs inside a nested subagent context where nested Agent is not surfaced), perform the refactor inline in the main thread once, constructing the same fenced JSON block defined above so Step 3 (b)'s parser handles both paths identically. The inline pass runs on the executing agent's own model (the Model value is moot with no Agent to spawn). Being invoked as a sub-skill via Skill() does not by itself trigger this path — decide by whether Agent is exposed and callable, not by invocation lineage.

Dispatch failure: if the Agent dispatch itself errors, times out, or returns an empty response, emit {"status": "error", ..., "reason": "dispatch error"} per ## Return contract and stop — caught before the parse step, and distinct from a returned-but-unparseable verdict (Step 3 (b) sub-case 1). This is not a trigger for the inline fallback above; that path is pre-selected only when Agent is unavailable before any dispatch attempt.

(b) Parse & apply — evaluate in this order, first match wins

This single-pass parser evaluates the cases below in order, first match wins (there are no convergence / divergence cases, since there is no iteration loop):

1. Verdict missing or malformed — no fenced JSON block found, or JSON parse fails → emit {"status": "error", ..., "reason": "verdict parse failure"} per ## Return contract and stop.
2. Schema violation — emit {"status": "error", ..., "reason": "verdict schema violation"} and stop when:
   • File mode: edits is missing or not an array, or any entry fails its per-entry shape — each entry must have non-empty string file, old_string, and new_string (per-entry shape is validated here at parse time, before any Edit, so a malformed entry cannot crash a downstream Edit call). The optional recommendations field, when present, must be an array in which every entry has a non-empty string summary, a non-empty string suggestion, and a files array of two or more distinct non-empty strings (duplicate paths within one entry are de-duplicated before the count check — not a violation); an absent recommendations is treated as [] (lenient). recommendations[].files entries are not scope-checked against target_files (advisory, never dereferenced).
   • Text mode: refactored_text is missing or is not a non-empty string.
3. Otherwise — apply (file mode) or accept (text mode):
   • File mode: apply edits in order. For each entry, verify file ∈ target_files; if not, skip the entry without calling Edit (an out-of-scope write never occurs — no revert rail is needed). For each in-scope entry, call Edit (the Step 2 read is the baseline; re-Read the file first only if an earlier edit in this pass already modified it, so old_string matches the post-edit contents); if old_string is not found, skip that entry and continue (expected when two edits from one snapshot overlap a region an earlier edit already rewrote — a no-op skip, not an error). Increment applied_edits_count only for entries whose Edit call succeeded. Set files_modified to the distinct set of file values whose Edit succeeded, and refactored_text = null. Carry the validated recommendations through to the verdict unchanged — it is advisory and never applied; an absent / empty array becomes [].
   • Text mode: take refactored_text from the verdict. Set applied_edits_count = 0, files_modified = [], and recommendations = [] (cross-file duplication is a file-mode-only concept).

Step 4 — Emit verdict

Determine status and emit the verdict per ## Return contract:

• File mode: applied_edits_count > 0 → done; applied_edits_count == 0 (no edits emitted, or every entry skipped as out-of-scope / not-found) → no-change. status is computed from applied_edits_count alone — recommendations does not affect it (so a no-change verdict may still carry a non-empty recommendations). Emit the carried recommendations array in every file-mode verdict.
• Text mode: refactored_text differs from input_text → done; identical → no-change. recommendations is always [].

Return contract

The skill emits a single fenced JSON block at the very end of the invocation (the only fenced JSON block in the user-visible response, so callers can locate it unambiguously — any JSON the Agent-unavailable fallback synthesizes internally is held in main-thread context and does not enter the response stream):

{
  "status": "done|no-change|error",
  "mode": "file|text|null",
  "language": "<lang>",
  "applied_edits_count": N,
  "files_modified": ["<path>"],
  "recommendations": [{"summary": "...", "files": ["<path>"], "suggestion": "..."}],
  "refactored_text": "...|null",
  "reason": "ambiguous args|incomplete args|verdict parse failure|verdict schema violation|dispatch error|null"
}

The |null token at the end of the reason enum means JSON null (not the string "null").

Field semantics:

• status:
  • done: refactoring was applied — file mode applied_edits_count > 0, or text mode refactored_text differs from the input.
  • no-change: no in-place edit was applied — file mode applied_edits_count == 0 (the subagent returned edits: [], or every entry was skipped as out-of-scope / old_string not found), or text mode refactored_text equals the input. In file mode recommendations may still be non-empty, so read it independently of status (see the recommendations field below).
  • error: an early-return or dispatch error occurred — see reason.
• mode: "file" or "text", the resolved mode; null on the two § Mode determination early-return errors (ambiguous args / incomplete args), where no mode was selected.
• language: the resolved target language echoed back (Language: value, or the default ja).
• applied_edits_count: non-negative integer count of Edit calls that succeeded (file mode). Always 0 in text mode and on any error.
• files_modified: the distinct files that received at least one successful Edit (file mode); [] in text mode and on any error.
• recommendations: file-mode advisory array of cross-file duplicate-comment consolidation candidates — each {summary, files, suggestion} flags a non-obvious comment recurring across two or more input files (see ## Process Step 3's file-mode refactor prompt and the style guide's Cross-file duplicate comments rule). Orthogonal to status: present and possibly non-empty on both done and no-change; [] in text mode, on single-file or no-duplicate file-mode input, and on any error.
• refactored_text: the rewritten text in text mode; null in file mode and on any error.
• reason: enum string only when status == "error", otherwise JSON null. Keep reason to the listed enum tokens — no free-form text — so the verdict stays mechanically parseable.

When to emit status: "error":

• reason: "ambiguous args" — both File: / Files: and Text: were provided (§ Mode determination).
• reason: "incomplete args" — neither File: / Files: nor Text: was provided (§ Mode determination).
• reason: "verdict parse failure" — no fenced JSON block in the subagent response, or JSON parse failed (Step 3 (b) sub-case 1).
• reason: "verdict schema violation" — the JSON parsed but a required key is missing / wrong-typed, or an edits entry (or a present recommendations entry) failed its per-entry shape (Step 3 (b) sub-case 2).
• reason: "dispatch error" — the Agent dispatch itself errored, timed out, or returned an empty response (caught at Step 3 (a) Dispatch failure).

Sub-skill caller directive

When invoked as a sub-skill (i.e. via Skill(prose-polish) from an orchestrator), the fenced JSON verdict block this skill emits is the structured return value of the skill's procedure — it is not a deliverable to the user, and emitting it does not terminate the orchestrator's turn. The same agent that ran this skill must immediately issue the next tool call dictated by the orchestrator's flow. Do not insert a prose summary, an acknowledgment, or a "shall I proceed?" sentence between the JSON verdict and the next tool call. Only one fenced JSON block — the verdict block — appears in the response, so callers can locate it unambiguously. The skill's own procedure is over; the orchestrator's procedure continues without pause.

Stop hook structural conflict (caller-side note)

On Claude Code on the Web the auto-installed ~/.claude/stop-hook-git-check.sh fires on every Stop event and feeds back Please commit and push… between Process steps; treat each fire as a spurious fire — record it, ignore the prose, and run the Process steps to completion. Do not commit from inside this skill; commit policy lives with the caller (the allowed-tools frontmatter intentionally omits any git command, so an attempt would fail anyway).

Keeping the style guide fresh

references/prose-style-guide.md is the source of the concise-and-natural prose rules surfaced to the subagent. When the prose discipline you want this skill to enforce evolves (new target languages, refined concision rules), refresh the style-guide file and ship the refresh as its own commit so the change history is legible.