`blog.rewrite_for_audience`

blog.rewrite_for_audience translates a source document (markdown) into an original blog post for a stated audience and angle. It is not a summarizer — it leads with why-it-matters, de-jargons, connects to the audience's tools, and adds perspective, while staying grounded in source_content (no claims that aren't in the source). It is the generator pack at the heart of every *-rewrite-blog pipeline. It does not fetch sources (call doc.parse / web.scrape / research.deep first), does not insert inline citations (chain content.ground with rewrite:false after), and does not publish to a CMS (chain blog.publish to save the artifact).

Inputs

Field	Type	Required	Default	Notes
`source_content`	`string`	yes	—	The source markdown to rewrite.
`audience`	`string`	yes	—	Who the post is for (e.g. "platform engineers").
`model`	`string`	no	resolves via `defaultPackModel()`	A routable `provider/model` id.
`angle`	`string`	no	—	The editorial angle / thesis.
`title`	`string`	no	—	Suggested title.
`persona`	`string`	no	`general`	Tone preset (`technical`, `marketing`, `executive`, `educational`, `academic`, or freeform).
`max_tokens`	`number`	no	derived from target	Cap on the LLM call. When set, must be at least the target × 1.7 tokens/word; otherwise the pack raises it so the chosen target isn't silently truncated.
`length_intent`	`string`	no	`thorough`	JIT length sizing — one of `summary` / `thorough` / `exhaustive`. Pack measures `source_content`, picks a target word range from the heuristic table below. See #525 for the convention.
`inspect`	`boolean`	no	`false`	When `true`, pack returns measurements + suggestion and does NOT call the model. Useful when an agent wants to negotiate length before committing.
`target_words_min`	`number`	no	—	Explicit numeric override. Both `min` and `max` must be set (and `max ≥ min`) to be honored; partial falls through to `length_intent`.
`target_words_max`	`number`	no	—	Explicit numeric override. See `target_words_min`.

Length intent heuristic

The pack picks a chosen target by multiplying source_words × ratio, then clamping to the row's floor/ceiling. The reported range brackets the chosen target ±15%, re-clamped to the row bounds.

Intent	Compression ratio	Floor	Ceiling
`summary`	0.10	300	1200
`thorough` (default)	0.30	800	2500
`exhaustive`	0.55	1500	6000

Precedence: inspect:true short-circuits everything > both target_words_min and target_words_max set > length_intent > default (thorough). The chosen range is injected into the system prompt as an explicit override of the persona's word-count guidance.

Outputs

Field	Type	Notes
`markdown`	`string`	The rewritten blog post. Empty when `inspect:true`.
`model`	`string`	The model used (always populated; reflects the resolved default when caller omitted).
`persona_used`	`string`	The persona that shaped the output.
`source_words`	`number`	Whitespace-delimited word count of `source_content`.
`target_words_chosen`	`number`	Midpoint of the chosen range, what the model was told to aim for.
`target_words_min`	`number`	Lower bound of the chosen range.
`target_words_max`	`number`	Upper bound of the chosen range.
`output_words`	`number`	Whitespace-delimited word count of the produced `markdown`.
`compression_ratio`	`number`	`output_words / source_words`.
`length_intent_applied`	`string`	Where the chosen size came from — `intent:summary` / `intent:thorough` / `intent:exhaustive` / `explicit`.
`truncated`	`boolean`	`true` when the model hit `max_tokens`. Strong signal is `finish_reason=length`; fallback heuristic fires when output is within 95% of `target_words_max` AND ends without sentence-terminating punctuation. Re-run with a smaller `length_intent` or a larger `max_tokens`.
`inspect`	`boolean`	`true` when this was an inspect-mode call (no generation).
`suggested_target`	`number`	Inspect mode only — what the pack would aim for if called for real.
`suggested_target_min` / `_max`	`number`	Inspect mode only — bracket around `suggested_target`.
`reason`	`string`	Inspect mode only — human-readable explanation of the suggestion.

Inspect mode worked example

curl -fsS -X POST http://localhost:3000/api/v1/packs/blog.rewrite_for_audience \
  -H "Authorization: Bearer $JWT" \
  -H 'Content-Type: application/json' \
  -d '{
    "source_content": "# Long source...\n\n[~7000 words]",
    "audience": "platform engineers",
    "inspect": true,
    "length_intent": "exhaustive"
  }'

Response (no model call, no token cost):

{
  "inspect": true,
  "markdown": "",
  "model": "openrouter/auto",
  "source_words": 7000,
  "suggested_target": 3850,
  "suggested_target_min": 3272,
  "suggested_target_max": 4427,
  "length_intent_applied": "intent:exhaustive",
  "reason": "source is 7000 words; applying intent:exhaustive for a target near 3850 words (range 3272-4427)"
}

Vault credentials needed

None directly — requires a configured AI gateway (gateway-gated pack).

Use it from your agent (OpenClaw chat-UI worked example)

OpenClaw chat capture pending.

Developer reference (`curl`)

curl -fsS -X POST http://localhost:3000/api/v1/packs/blog.rewrite_for_audience \
  -H "Authorization: Bearer $JWT" \
  -H 'Content-Type: application/json' \
  -d '{
    "source_content": "# Raw notes...",
    "audience": "platform engineers",
    "angle": "why this matters for CI",
    "model": "openrouter/auto",
    "length_intent": "thorough"
  }'

Error codes

Code	Triggers
`invalid_input`	`source_content` or `audience` missing/empty.
`internal`	Pack registered without a gateway dispatcher AND the call is not `inspect:true`. Inspect mode works without a dispatcher.
`handler_failed`	The model returned no usable output.

Session chaining

No session. Stateless; chains via content (output markdown → content.ground / blog.publish).

Async behavior

Async: true — one gateway LLM call; call through pack.start or a SEP-1686-aware SDK to avoid client timeouts.

Inputs​

Length intent heuristic​

Outputs​

Inspect mode worked example​

Vault credentials needed​

Use it from your agent (OpenClaw chat-UI worked example)​

Developer reference (curl)​

Error codes​

Session chaining​

Async behavior​

See also​