fable-prompt · guide-aligned request rewrite (per-request layer)
Source: Anthropic docs "Prompting Claude Fable 5.1" (https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-1).
Fixed guide blocks: references/prompt-blocks.md. Before/after samples: references/examples.md.
This skill handles the per-request layer only: the four fields every request needs, plus the guide
blocks that depend on the request. The always-on rules (autonomy, scope limits, targeted edits, progress
updates, formatting, batching) are injected by this plugin's SessionStart hook, so never repeat them here.
Mode
- Default: rewrite, show, then execute in the same turn. Do not stop after showing the prompt.
- Show only when the user says "프롬프트만", "보여만 줘", "just the prompt", "don't run it".
- Never ask the user to write the prompt. Build it from conversation context.
Step 1 · Resolve referents
Fill "이거", "그거", "this", "that file", "the error" from context in this order: last path mentioned, last
error text, last artifact produced, current git diff, open thread. Write the resolved value as a concrete
path, symbol, or quote.
Ask exactly one question (with options and a recommended pick) only when different readings would lead
to materially different work. Routine ambiguity: pick the reading the wording and surrounding code
most directly support and state the assumption inside the prompt.
Step 2 · Classify
| Kind |
Signal |
Deliverable |
| Change |
build/fix/change verbs |
working change + verification evidence |
| Assessment |
user describes a problem, asks why, thinks out loud |
findings only, no fix until asked (guide exception) |
| Research |
look up, investigate, names of tools or models, anything time-sensitive |
sourced answer; search the name as the user wrote it |
| Writing |
write, summarise, draft a doc or post |
text in the requested shape, no mannered prose |
| Long deliverable |
full rewrite, multi-section doc, big table, whole file |
as above plus the long-output note (block G) |
Step 3 · Compose
Write task-specific parts (goal, context, scope, done) in the user's language so they can check them. Keep every guide block in English verbatim; never translate a block.
- Goal · one sentence, outcome-verifiable.
- Context · resolved paths, symbols, error text, related decisions.
- Scope · what is in, what is explicitly out.
- Done criteria · the exact check: a command, a count, a file that must exist, a reproduced workflow.
- Effort · do not write an effort line into the request; a line in a prompt changes nothing in Claude Code.
Effort is a setting (
effortLevel in settings.json or CLAUDE_CODE_EFFORT_LEVEL). Mention it only when it
matters for this request: time-sensitive research at low → add block H; a long deliverable at xhigh
→ add block G.
- Conditional blocks · Assessment → the "report findings, don't fix" sentence. Research/summary →
block J (quoting example). Writing → block F (short form). Code task → phrase checks as
"Are there any bugs?" not "Does it compile?" (safeguard false positives).
Never attach blocks A, B, C, D, E, or I. They are already active through the plugin hook. The improved
request stays short: four fields, effort, and only the conditional lines from item 6.
Step 4 · Show, then run
Print the prompt in one fenced block titled 개선된 요청 (or Improved request), four fields only plus any conditional lines, then execute it as if
the user had sent it. Open with one line on what you are doing, give brief updates, and close with a
recap that stands on its own. End with exactly one status: DONE, DONE_WITH_CONCERNS, BLOCKED, or NEEDS_CONTEXT.
Do not
- Do not widen the task while improving it. The rewrite clarifies; it does not add features.
- Do not turn an Assessment into a Change.
- Do not paraphrase block A's first paragraph; its opening sentence carries most of the effect.
- Do not add anti-formatting rules; Fable 5.1 already under-formats.
1---2name: fable-prompt3description: fable-prompt · guide-aligned request rewrite (per-request layer)4---5# fable-prompt · guide-aligned request rewrite (per-request layer)67Source: Anthropic docs "Prompting Claude Fable 5.1" (https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-1).8Fixed guide blocks: `references/prompt-blocks.md`. Before/after samples: `references/examples.md`.910This skill handles the **per-request layer** only: the four fields every request needs, plus the guide11blocks that depend on the request. The always-on rules (autonomy, scope limits, targeted edits, progress12updates, formatting, batching) are injected by this plugin's SessionStart hook, so never repeat them here.1314## Mode15- **Default: rewrite, show, then execute in the same turn.** Do not stop after showing the prompt.16- **Show only** when the user says "프롬프트만", "보여만 줘", "just the prompt", "don't run it".17- Never ask the user to write the prompt. Build it from conversation context.1819## Step 1 · Resolve referents20Fill "이거", "그거", "this", "that file", "the error" from context in this order: last path mentioned, last21error text, last artifact produced, current git diff, open thread. Write the resolved value as a concrete22path, symbol, or quote.2324Ask exactly one question (with options and a recommended pick) only when different readings would lead25to **materially different work**. Routine ambiguity: pick the reading the wording and surrounding code26most directly support and state the assumption inside the prompt.2728## Step 2 · Classify29| Kind | Signal | Deliverable |30|---|---|---|31| Change | build/fix/change verbs | working change + verification evidence |32| Assessment | user describes a problem, asks why, thinks out loud | findings only, **no fix** until asked (guide exception) |33| Research | look up, investigate, names of tools or models, anything time-sensitive | sourced answer; search the name as the user wrote it |34| Writing | write, summarise, draft a doc or post | text in the requested shape, no mannered prose |35| Long deliverable | full rewrite, multi-section doc, big table, whole file | as above plus the long-output note (block G) |3637## Step 3 · Compose38Write task-specific parts (goal, context, scope, done) in the user's language so they can check them. Keep every guide block in English verbatim; never translate a block.39401. **Goal** · one sentence, outcome-verifiable.412. **Context** · resolved paths, symbols, error text, related decisions.423. **Scope** · what is in, what is explicitly out.434. **Done criteria** · the exact check: a command, a count, a file that must exist, a reproduced workflow.445. **Effort** · do not write an effort line into the request; a line in a prompt changes nothing in Claude Code.45 Effort is a setting (`effortLevel` in settings.json or `CLAUDE_CODE_EFFORT_LEVEL`). Mention it only when it46 matters for this request: time-sensitive research at `low` → add block **H**; a long deliverable at `xhigh`47 → add block **G**.486. **Conditional blocks** · Assessment → the "report findings, don't fix" sentence. Research/summary →49 block **J** (quoting example). Writing → block **F** (short form). Code task → phrase checks as50 "Are there any bugs?" not "Does it compile?" (safeguard false positives).5152**Never attach blocks A, B, C, D, E, or I.** They are already active through the plugin hook. The improved53request stays short: four fields, effort, and only the conditional lines from item 6.5455## Step 4 · Show, then run56Print the prompt in one fenced block titled `개선된 요청` (or `Improved request`), four fields only plus any conditional lines, then execute it as if57the user had sent it. Open with one line on what you are doing, give brief updates, and close with a58recap that stands on its own. End with exactly one status: DONE, DONE_WITH_CONCERNS, BLOCKED, or NEEDS_CONTEXT.5960## Do not61- Do not widen the task while improving it. The rewrite clarifies; it does not add features.62- Do not turn an Assessment into a Change.63- Do not paraphrase block A's first paragraph; its opening sentence carries most of the effect.64- Do not add anti-formatting rules; Fable 5.1 already under-formats.
Junhan2