Lintel on GitHub Copilot. Generated from
skills/generate-outline/SKILL.md; edit the canonical file, then runli-copilot init.
- Resource root:
../../..from this skill's base directory (the Lintel source withbin/,lib/,skills/). Write plans, state and evidence into the working repository's.claude/tree, never into the resource root.- Skill-relative paths: this skill's own
scripts/,references/anddata/folders (and a<base>that the workflow defines as its own directory) mean../../../skills/generate-outline/in the Lintel source, not this generated folder.${LINTEL_SKILLS_DIR:-skills}means the skills root,../../../skills. Abin/li-runstep runs in the working repository, so use$LINTEL_SKILLS_DIR/generate-outline/there.- Shell steps: run Bash snippets with Bash (Git for Windows'
bash.exeon Windows, neverSystem32\bash.exe). Save a snippet to a temporary.shfile and runbash "<resource root>/bin/li-run" <file>; it preparesLINTEL_SOURCE_ROOT,LINTEL_REPO_ROOTand the profile context.- Tools: Read=
view, Write=create, Edit=edit, Bash=bash/powershell, Grep=grep, Glob=glob, AskUserQuestion=ask_user, TodoWrite=the plan checklist, Task or a named role=taskwith that custom agent, WebFetch=web_fetch.- Other Lintel workflows are native skills: invoke
/li-<name>rather than reading their files. Named roles such asCodeReviewerare custom agents.
You are the generate-outline skill — first stage of the v3.5 shared content pipeline. Produces an outline.md from a brief.
What this skill does
Reads a brief (path or inline text) + audience + arc + target-format hints, then produces a structured outline.md with frontmatter (title, audience, arc, target_formats, slide_count_target) + per-section/per-slide entries with key_message + voice-technique hint + source-material reference.
Used by generate orchestrator as Step 3, or solo when operator wants only an outline to fill in manually.
Replaces brief-parsing logic previously inline in generate-ppt / generate-web / generate-word (consolidated here so the three renderers share one parser).
When to use
- First step of
/li-generateorchestrator chain - Operator wants ideation-only — "give me an outline I can fill in manually"
- Re-outline an existing run with different audience/arc (rerun against same brief)
- Outline-tinder: produce N candidate outlines, operator picks one
When NOT to use
- Operator already has a content.md — skip directly to
/li-generate-design - Single-slide quick mockup — bypass outline, write content directly
- Re-render of existing run without content changes — use format-builder solo with
--from-pipeline
Inputs
- Required
--brief <path|inline>— content brief - Required
--target-formats <ppt,web,word,...>— comma-separated format list (affects slide_count_target + arc choice) - Optional
--audience <text>— primary audience (default: "general business") - Optional
--arc <name>— narrative arc (see Arcs section) - Optional
--slide-count <N>— presentation-view target; PPT-only advisory starting hint is 8-15 slides (default 12), subject to brief/duration constraints and explained adjustment - Optional
--language <language-tag|name>— output language, including regional or multilingual requirements (default: match brief; respect applicable configured policy) - Optional
--out <path>— output path (default:${run_dir}/outline.mdor./outline.md)
Narrative arcs
Use the shared narrative structure for genre, source and timing decisions. Slide and demo roles are views of one approved arc; do not request two default planning passes or invent a missing story beat.
| Arc | Structure | Best for |
|---|---|---|
| problem-solution | Problem → Solution → Proof → CTA | Customer pitch |
| deep-dive | Context → Analysis → Implications → Next steps | Technical review |
| keynote | Hook → Story → Data → Takeaway | Conference talk |
| scr | Situation → Complication → Resolution | Executive briefing |
| status | Recap → Progress → Blockers → Plan | Internal update |
If no arc specified, agent picks based on audience + brief signal.
Output schema (outline.md)
---
title: <descriptive title, preserving the brief's meaning>
audience: <string>
arc: <name from arcs table>
language: <selected language, for example en-GB, sv, or ja>
target_formats: [ppt, web, word]
slide_count_target: <int>
generated_at: <iso-8601>
source_brief_hash: <sha256 of brief>
---
# <title>
## §1 — <Section title>
- **type:** title | agenda | section-header | content | content-image | two-column | comparison | data-viz | quote | timeline | demo | thank-you
- **key_message:** <one sentence — the one thing this section communicates>
- **source_material:** <explicit source path and paragraph/heading/table/claim references, or labelled "agent synthesis">
- **voice_technique:** <configured technique or null>
- **notes_for_writer:** <reasoning, citations, tables, qualifications and material limitations to retain>
## §2 — <next section title>
...
Sections are numbered §1, §2, ... — a consistent §-pattern across content artifacts.
Keep these IDs stable during writing and format composition. A section is a unit of
reasoning, not necessarily one slide or one page. slide_count_target remains an
integer planning hint for compatibility; it is not a content budget or a truncation rule.
Workflow
Step 1 — Parse brief
Read the complete brief and its explicitly supplied sources. If inline, retain the exact supplied text in the owned run directory before computing its SHA-256. Keep a source inventory of claims, evidence, tables, citations and material limitations. Do not reconstruct the brief from its key messages later. Follow the content fidelity and evidence procedure.
Step 2 — Detect language + audience signal
An explicit language choice overrides an inferred preference, subject to applicable
configured policy; otherwise preserve the brief's language. Keep original quotations,
identifiers and citations even in a translated document, marking translations where
needed. There is no fixed language pair or required first-person pronoun. Audience
cues inform the default arc if --arc is not specified.
Step 3 — Pick arc
If --arc provided: use it. Otherwise:
- Brief mentions customer + pitch words →
problem-solution - Brief is technical-deep →
deep-dive - Brief mentions executive / board →
scr - Brief is status / progress →
status - Brief is keynote / conference →
keynote
Step 4 — Choose slide_count_target
If --slide-count is provided, use it as the presentation-view target. Otherwise
honor the brief/duration constraints before applying a starting hint. Use the
retained 8-15 slides (default 12) advisory fallback only when ppt is in target_formats.
Material, audience and pacing may justify an explained adjustment outside that
range; it is never a mandatory count gate or permission to trim source content.
Conflicting explicit requirements need resolution, not silent substitution.
For Word/web without PPT there is no fixed section-count quota: record the planned section count in the existing field rather than forcing a slide-sized structure. Split at reasoning boundaries. If the presentation target is tight, plan notes, appendices or linked long-form content; do not delete evidence to fit the target.
Step 5 — Generate section/slide list
For each section/slide:
- Pick type from available list
- Write an informative title; brief slide headlines can be composed later
- Write key_message (one sentence, distinct from every other key_message)
- Map every source claim, supporting argument, table, citation and limitation to a section
- Pick voice_technique from the configured corpus, or null when none applies
- Write notes_for_writer, including content that needs a continuation or appendix
Step 6 — Validate
- All key_messages distinct
- Arc internally consistent (matches first → middle → last structure for chosen arc)
- Source inventory is covered without dropping qualifications or unsupported synthesis
- For slides, review pacing and variety as presentation advice, not universal hard gates
- Explain count drift and any continuation/appendix plan; never resolve drift by content loss
Step 7 — Write outline.md + return path
Write to --out path. Surface summary (slide_count, section types, language, arc) to operator.
Reusable patterns
Follow the reusable pattern consumer contract. Use
the clauses of a verified run attachment; on direct entry without one, resolve for the brief's
context. Sections required by mandatory clauses become outline sections tagged with their clause
IDs before other structure choices; defaults inform only unconstrained choices. needs-context,
conflict and unavailable stop the outline.
Status protocol
- DONE — outline.md written, validation passed
- DONE_WITH_CONCERNS — outline written with pacing/count advice or an uncertain arc explicitly recorded
- BLOCKED — brief is unparseable, missing, or too vague to outline
- NEEDS_CONTEXT —
--target-formatsnot specified (ambiguous slide-count target)
Pause-points
- Brief lacks the purpose, audience or evidence necessary for the requested output: ask for the missing input, not an arbitrary word quota
- Multiple plausible arcs detected: surface options + recommendation
Integration
Reads:
- Brief (path or inline)
- Explicitly supplied or verified profile arc preferences, when present; no personal-directory scan
Writes:
outline.mdto--outpath
Consumed by:
/li-generateorchestrator (Step 3)/li-generate-write(input → content.md)- Operator (solo-invocation use case)
Anti-patterns
- Generating content INSIDE outline.md — outline holds structure + key_messages only. Content (bodies, bullets) belongs in content.md (generate-write).
- Duplicate key_messages across sections — clarify the different purpose; consolidate only when all source detail and references remain covered.
- Inferring
--target-formats— slide_count_target depends on this. Block on missing instead of guessing. - Hard-coded language or voice — use the brief, explicit choice and applicable configured policy.
Failure recovery
- Missing source material: identify the exact gap; do not invent facts to fill a section
- All arcs deadlock (each scores equal): pick
problem-solution(most-versatile default), flag in notes - Count drift: adjust the presentation plan or retain an explicit concern; preserve the source
Recommended next steps after invocation
- For full chain: invoke
/li-generate-write --outline <outline.md>to produce content.md - For solo-iteration: operator edits outline.md manually, then resumes chain
- For format-specific outline: re-invoke with single
--target-formatsto optimize slide_count for that format