Lintel on GitHub Copilot. Generated from skills/generate-ppt/SKILL.md; edit the canonical file, then run
li-copilot init.
- Resource root:
../../.. from this skill's base directory (the Lintel source with bin/,
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/ and data/ folders (and a
<base> that the workflow defines as its own directory) mean
../../../skills/generate-ppt/ in the Lintel source, not this generated folder.
${LINTEL_SKILLS_DIR:-skills} means the skills root, ../../../skills. A bin/li-run step
runs in the working repository, so use $LINTEL_SKILLS_DIR/generate-ppt/ there.
- Shell steps: run Bash snippets with Bash (Git for Windows'
bash.exe on Windows, never
System32\bash.exe). Save a snippet to a temporary .sh file and run
bash "<resource root>/bin/li-run" <file>; it prepares LINTEL_SOURCE_ROOT, LINTEL_REPO_ROOT
and 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=task
with that custom agent, WebFetch=web_fetch.
- Other Lintel workflows are native skills: invoke
/li-<name> rather than reading their
files. Named roles such as CodeReviewer are custom agents.
/generate-ppt
Produces an editable PowerPoint deck (.pptx) from a complete source brief. Visible
slides are a presentation view, not permission to discard evidence or reasoning.
Apply the selected profile's actual brand/voice requirements without inventing
neutral thresholds or scanning personal templates. Generation is not distribution.
Prerequisites
Discover the actual host tool schemas first. Prefer an available native slide API:
on the Copilot canvas surface call list_canvas_capabilities for powerpoint,
then open_canvas on a new owned path and use discovered get_model, batch,
set_notes and render_slide actions. Follow the
native PowerPoint procedure.
The existing declared Node alternative is pptxgenjs. Inspect its task-local
availability and supported operations before use; do not assume it imports an
arbitrary .pptx template. Record a missing-tool failure before considering an
authorized task-local restore. No automatic/global install or security bypass.
If no editable writer exists, retain the full source and mark PPTX blocked.
HTML/PDF/Markdown may be explicitly chosen alternatives, never silently relabelled
editable PowerPoint or rendered PPTX evidence.
When to use
- Customer-engagement deliverable (pitch deck, workshop deck, summary deck)
- Internal presentation that needs the active pack's brand identity
- Reusing engagement-specific content across multiple PPT outputs
- Replacing manual PPT authoring with a workflow that has voice + brand + provenance gates
When NOT to use
- Static UI mockup rather than a slide — use
/generate-web --mode mockup
- Non-PPT artifact — use
/generate-word or /generate-web
- Pack compliance gate not satisfied AND output is customer-bound — resolve the gate first
Inputs
- Required
--brief <path|inline> — content brief describing the deck purpose OR --from-pipeline <dir> (shared pipeline mode)
- Required
--template <name|path> unless --use-defaults is selected — resolve names only within the verified configured template directory
- Optional
--audience <text> — primary audience (affects voice tier output)
- Optional
--slide-count <N> — presentation-view target; retained advisory fallback is 20-30 slides based on duration, not a source-content cap
- Optional
--duration <minutes> — presentation duration (informs slide pacing)
- Optional
--voice <tier> — voice tier for slide content (default: the active pack's voice tier, internal by default)
- Optional
--use-defaults — available neutral template or explicitly blank native deck; do not invent a bundled template
- Optional
--ignore-stale-brand <reason> — advisory exception only; does not waive mandatory policy
- Optional
--out <path> — new .pptx output (default: <brief-stem>.pptx in the working directory)
From-pipeline mode (v3.5 Phase 2 — generate-pipeline integration)
If invoked with --from-pipeline <run-dir> instead of --brief:
Preserve this entry point and its existing fields. First follow
existing pipeline input admission
with --format ppt, external input context, original package/leaves and current
profile. It verifies canonical content/notes/hash identity and the existing PPT
projection without rendering. Actual artifact/source-retention and required P05
inspections remain separate; retired document-format acceptance is not a live
gate. Standalone --brief still needs no generate-design output or new domain envelope.
Read shared pipeline-output:
<run-dir>/content.md — written content (with HTML-comment annotations for voice/type/key_message per section)
<run-dir>/speaker-notes.md — speaker notes per slide
<run-dir>/design-spec.json — per-format layout-mappings (read per_format.ppt.layouts)
Replace brief-parsing logic with direct-read of content.md sections +
design-spec layouts. Do not run a second narrative rewrite that could discard
shared content; the format-specific fidelity review in step 3 still applies.
Apply format-specific design-pass via design_pass_hook:
- Reads
per_format.ppt.layouts[N].design_pass_hook (canonical: PPTNarrativeArchitect)
- Invokes agent for a PPT-specific fidelity-pass (slide-narrative-arc, FastPath-recommendations, animation-cues) on top of the shared baseline
- Keep format-specific design methods separate from shared content generation
CLI stays backward-compat: existing --brief-flag invocations work unchanged. --from-pipeline is additive.
Use the same source-retention, native inspection and configured controls as
standalone mode. Layout mappings must not discard unmapped source detail.
Keep complete returned content and notes, including facts not shown in a
visible placeholder. Continuation layouts keep the original section ID.
Select template/config overrides before admission and refresh final P05
evidence after the deck is created or edited; input checks do not establish
notes serialization, editability or rendered layout.
Existing skill arguments and data field names are retained; this is not a claim
that every host or shared-pipeline script has been executed.
Workflow
Preflight and policy:
- Follow the P05/P07 evidence procedure.
- Select explicit source/output/template roots; refuse unapproved replacement.
- Verify the pinned profile, actual template capability and requested inspections.
- Derive mandatory versus advisory brand, voice, freshness and disclosure controls
from applicable configuration and the brief, not a fixed score or age proxy.
- A required missing tool/control remains blocked; neutral mode needs no logo.
Read brief + extract structure:
- Goal / key message (one sentence)
- Audience profile
- All substantive sections, claims/evidence, tables, citations and material limitations
- Opening context and closing decision/CTA where appropriate
- Retain full source separately from the concise slide view
Standalone count selection: Explicit --slide-count takes precedence.
Honor brief/duration constraints before the retained 20-30 slides based on
duration starting range; material, audience and pacing can justify an
explained adjustment. This is an advisory fallback, not a mandatory gate,
not a Word/web section quota, and not the outline stage's 8-15/default-12 hint.
Surface conflicting explicit requirements without deleting source content.
2b. Design DNA slide pass (ADR-0017 — retrieval before slide design). Query the slide
decision engine so the arc is grounded in the corpus, not invented:
dna="${LINTEL_SOURCE_ROOT:?select the trusted source}/skills/design-dna"
python3 "$dna/scripts/search.py" "<deck goal / pitch type>" --slide strategy -n 1 # narrative arc + sparkline-beats
# then per slide, by the slide's emotion (trust|urgency|confidence…) and goal (hook|proof|cta…):
python3 "$dna/scripts/search.py" "<emotion>" --slide color-logic -n 1 # background/text/accent + full-bleed
python3 "$dna/scripts/search.py" "<goal>" --slide layout-logic -n 1 # layout pattern + break-pattern
python3 "$dna/scripts/search.py" "<slide-type>" --slide copy -n 1 # headline formula
Feed the strategy's sparkline_beats + emotion_arc to PPTNarrativeArchitect as the arc spine.
python3 absent → Read design-dna/data/slides/*.csv directly (controlled vocabulary in design-dna SKILL.md).
Use the accepted read-only PPTNarrativeArchitect method to design the arc;
delegate only through a real available host operation, otherwise label the
builder's own pass as self-review:
- Slide-by-slide content goals — seeded by the slide-strategy
sparkline_beats
- Mode tags from the configured corpus when applicable; neutral is valid
- Layout suggestions per slide — from
--slide layout-logic (pattern + break-pattern at 1/3, 2/3)
- Asset suggestions (from the active pack's asset library, if one is configured)
Generate editable slides through the selected native API or pptxgenjs:
- Apply an explicit compatible template or the requested neutral blank design
- Add slides per architect's arc
- Compose readable visible text without changing the scope of source claims
- Insert assets from the active pack's asset library where matched
- Write complete supporting paragraphs, citations, tables and the claim ledger
into actual speaker notes, an appendix or delivered linked long-form content.
Always retain the full source. Put a material limitation on the visible slide
whenever omitting it would misrepresent a visible claim.
- Keep source §N IDs and record continuation/appendix locations. A slide-count
target cannot authorize losing facts, shrinking illegibly or clipping text.
4b. Critique narration using the accepted read-only SlideNarrationCritic
method when a talk track exists. Supply the exact deck/notes, intended audience,
duration and current profile reference through the available host delegation
operation. Check pacing and recovery lines without deleting supporting evidence.
Keep speaker talk time separate from appendix/reference material. A claimed
words-per-minute target is not an observed rehearsal.
Inspect the saved artifact, then apply the familiar control categories:
Run the shipped read-only check_pptx.py source-retention procedure
on the actual deck and explicit source paragraphs/cells. It checks bounded
ZIP/XML and real note relationships; keep missing detail and malformed/aliased
notes as findings. This does not supply application or rendered-layout evidence.
First reopen through the actual available application/API, make a small scoped
edit and read it back. Compare full source to slides plus actual saved notes.
Render every slide using the available renderer and inspect wrapping, overflow,
overlap, table legibility and assets. Check notes separately; slides alone cannot
prove note retention. Fix layout by reflow/splitting, not by discarding facts.
Voice: apply only the verified corpus's requirements and applicability.
Neutral clarity/pacing advice does not become a hard vocabulary or score gate.
Brand: check configured template/tokens and explicitly authorized local
assets. No invented logo, personal library scan or claim that a blank deck is
company-branded. Missing mandatory brand data cannot silently fall back.
Use the named brand-source and brand-freshness
selected asset evidence procedures.
Record actual source/version and refresh requirements as P05 check controls;
unknown currency stays unverified. --ignore-stale-brand records only a
permitted advisory exception, not a passed mandatory freshness check.
Honest limitations: check each material claim against its evidence,
assumptions and failure boundary, including claims on pitch/workshop slides.
Disclosure requirements follow the brief/policy; counts cannot prove honesty.
Record a grounded N/A only for a genuinely inapplicable requirement.
Provenance: bind the exact brief, source content, notes, template/config,
output and live P07 reference using existing P05 snapshot/control evidence.
File metadata or a made-up provenance ID cannot replace those observations.
Report the actual result at the explicit owned output path.
Mandatory failure/error/unverified blocks the affected acceptance or
distribution regardless of scores. Retain useful editable output and its
source while naming any missing renderer or review. Do not use an implicit
personal draft directory. An API edit/readback is not every-client acceptance.
Reusable patterns
Follow the reusable pattern consumer contract.
Brief mode resolves for the brief's context; from-pipeline mode verifies the run's
pattern_context attachment. Both receive the same clauses before the slide structure is
chosen: required sections become slides or sections, and the clauses go to the existing
narrative review and the QA report. A helper fixture proves only this integration, not a real
generated deck.
Report format
Record exact source/output paths and hashes; actual profile/template; audience and
arc; slide-to-section/notes mapping; writer and native provider/instance/actions;
edit/readback result; rendered slides inspected and specific renderer limits;
retained claims, citations, table data and material limitations; configured
control outcomes; and remaining acceptance/review gates. Keep documentary checks,
package extraction, native actions and render evidence separate.
Compliance integration
- Configured applicable controls keep their mandatory/advisory semantics.
- No customer data, secrets, macros, external upload or automatic asset download
in synthetic validation. Use explicit authorized local assets.
- Required independent review and delivery controls remain separate from creation.
Voice tier note
voice: mixed. Skill itself is engineering-internal; the SLIDE CONTENT may use a customer-facing voice tier (gated by Gate 1).
Failure modes
- Writer error — preserve source and the owned draft, record the actual error; do not emit a passing gate.
- Required template missing — block; offer neutral defaults only when applicable policy permits.
- Voice gate fails after 2 regen attempts — keep draft, surface specific slide failures with fix recommendations
- Required asset unavailable — report its exact gap; do not invent a successful asset inspection.
- Renderer absent — retain editable deck/source but keep rendered-layout acceptance unverified.
- Customer-data in brief — BLOCK before generation. Sanitize first.
Examples
Standard pitch deck:
> /generate-ppt --brief docs/engagement/customer-A-pitch-brief.md --template pitch-deck --audience "public-sector CIO"
[Resolve the explicit template, preserve evidence, and inspect the actual artifact.]
Workshop deck with defaults:
> /generate-ppt --brief synthetic-brief.md --template workshop --use-defaults --voice internal --out workshop.pptx
[An explicitly blank native deck is valid neutral output; configured requirements still apply.]
Stale brand override:
> /generate-ppt --brief archive-pitch.md --template pitch-deck --ignore-stale-brand "archival deliverable, brand version pinned"
[Records an advisory exception; does not waive required brand policy.]
See also
1---2name: li-generate-ppt3description: Use to create an editable PowerPoint deck from a brief or shared pipeline, retaining source detail in notes and inspecting actual rendered slides.4metadata:5 source: https://skillmd.com/skills/jokerman89/li-generate-ppt6---78> **Lintel on GitHub Copilot.** Generated from `skills/generate-ppt/SKILL.md`; edit the canonical file, then run9> `li-copilot init`.10> - **Resource root:** `../../..` from this skill's base directory (the Lintel source with `bin/`,11> `lib/`, `skills/`). Write plans, state and evidence into the working repository's `.claude/`12> tree, never into the resource root.13> - **Skill-relative paths:** this skill's own `scripts/`, `references/` and `data/` folders (and a14> `<base>` that the workflow defines as its own directory) mean15> `../../../skills/generate-ppt/` in the Lintel source, not this generated folder.16> `${LINTEL_SKILLS_DIR:-skills}` means the skills root, `../../../skills`. A `bin/li-run` step17> runs in the working repository, so use `$LINTEL_SKILLS_DIR/generate-ppt/` there.18> - **Shell steps:** run Bash snippets with Bash (Git for Windows' `bash.exe` on Windows, never19> `System32\bash.exe`). Save a snippet to a temporary `.sh` file and run20> `bash "<resource root>/bin/li-run" <file>`; it prepares `LINTEL_SOURCE_ROOT`, `LINTEL_REPO_ROOT`21> and the profile context.22> - **Tools:** Read=`view`, Write=`create`, Edit=`edit`, Bash=`bash`/`powershell`, Grep=`grep`,23> Glob=`glob`, AskUserQuestion=`ask_user`, TodoWrite=the plan checklist, Task or a named role=`task`24> with that custom agent, WebFetch=`web_fetch`.25> - **Other Lintel workflows** are native skills: invoke `/li-<name>` rather than reading their26> files. Named roles such as `CodeReviewer` are custom agents.2728# /generate-ppt2930Produces an editable PowerPoint deck (.pptx) from a complete source brief. Visible31slides are a presentation view, not permission to discard evidence or reasoning.32Apply the selected profile's actual brand/voice requirements without inventing33neutral thresholds or scanning personal templates. Generation is not distribution.3435## Prerequisites3637Discover the actual host tool schemas first. Prefer an available native slide API:38on the Copilot canvas surface call `list_canvas_capabilities` for `powerpoint`,39then `open_canvas` on a new owned path and use discovered `get_model`, `batch`,40`set_notes` and `render_slide` actions. Follow the41[native PowerPoint procedure](../../../skills/generate-ppt/references/native-powerpoint.md).4243The existing declared Node alternative is `pptxgenjs`. Inspect its task-local44availability and supported operations before use; do not assume it imports an45arbitrary `.pptx` template. Record a missing-tool failure before considering an46authorized task-local restore. No automatic/global install or security bypass.47If no editable writer exists, retain the full source and mark PPTX blocked.48HTML/PDF/Markdown may be explicitly chosen alternatives, never silently relabelled49editable PowerPoint or rendered PPTX evidence.5051## When to use5253- Customer-engagement deliverable (pitch deck, workshop deck, summary deck)54- Internal presentation that needs the active pack's brand identity55- Reusing engagement-specific content across multiple PPT outputs56- Replacing manual PPT authoring with a workflow that has voice + brand + provenance gates5758## When NOT to use5960- Static UI mockup rather than a slide — use `/generate-web --mode mockup`61- Non-PPT artifact — use `/generate-word` or `/generate-web`62- Pack compliance gate not satisfied AND output is customer-bound — resolve the gate first6364## Inputs6566- Required `--brief <path|inline>` — content brief describing the deck purpose **OR** `--from-pipeline <dir>` (shared pipeline mode)67- Required `--template <name|path>` unless `--use-defaults` is selected — resolve names only within the verified configured template directory68- Optional `--audience <text>` — primary audience (affects voice tier output)69- Optional `--slide-count <N>` — presentation-view target; retained advisory fallback is 20-30 slides based on duration, not a source-content cap70- Optional `--duration <minutes>` — presentation duration (informs slide pacing)71- Optional `--voice <tier>` — voice tier for slide content (default: the active pack's voice tier, `internal` by default)72- Optional `--use-defaults` — available neutral template or explicitly blank native deck; do not invent a bundled template73- Optional `--ignore-stale-brand <reason>` — advisory exception only; does not waive mandatory policy74- Optional `--out <path>` — new `.pptx` output (default: `<brief-stem>.pptx` in the working directory)7576## From-pipeline mode (v3.5 Phase 2 — generate-pipeline integration)7778If invoked with `--from-pipeline <run-dir>` instead of `--brief`:7980Preserve this entry point and its existing fields. First follow81[existing pipeline input admission](../../../skills/generate-write/references/fidelity-and-evidence.md#existing-pipeline-input-admission)82with `--format ppt`, external input context, original package/leaves and current83profile. It verifies canonical content/notes/hash identity and the existing PPT84projection without rendering. Actual artifact/source-retention and required P0585inspections remain separate; retired document-format acceptance is not a live86gate. Standalone `--brief` still needs no generate-design output or new domain envelope.87881. **Read shared pipeline-output:**89 - `<run-dir>/content.md` — written content (with HTML-comment annotations for voice/type/key_message per section)90 - `<run-dir>/speaker-notes.md` — speaker notes per slide91 - `<run-dir>/design-spec.json` — per-format layout-mappings (read `per_format.ppt.layouts`)92932. **Replace brief-parsing logic** with direct-read of content.md sections +94 design-spec layouts. Do not run a second narrative rewrite that could discard95 shared content; the format-specific fidelity review in step 3 still applies.96973. **Apply format-specific design-pass via design_pass_hook:**98 - Reads `per_format.ppt.layouts[N].design_pass_hook` (canonical: PPTNarrativeArchitect)99 - Invokes agent for a PPT-specific fidelity-pass (slide-narrative-arc, FastPath-recommendations, animation-cues) on top of the shared baseline100 - Keep format-specific design methods separate from shared content generation1011024. **CLI stays backward-compat:** existing `--brief`-flag invocations work unchanged. `--from-pipeline` is additive.1031045. **Use the same source-retention, native inspection and configured controls as105 standalone mode.** Layout mappings must not discard unmapped source detail.106 Keep complete returned content and notes, including facts not shown in a107 visible placeholder. Continuation layouts keep the original section ID.108 Select template/config overrides before admission and refresh final P05109 evidence after the deck is created or edited; input checks do not establish110 notes serialization, editability or rendered layout.111112Existing skill arguments and data field names are retained; this is not a claim113that every host or shared-pipeline script has been executed.114115## Workflow1161171. **Preflight and policy:**118 - Follow the [P05/P07 evidence procedure](../../../skills/generate-write/references/fidelity-and-evidence.md).119 - Select explicit source/output/template roots; refuse unapproved replacement.120 - Verify the pinned profile, actual template capability and requested inspections.121 - Derive mandatory versus advisory brand, voice, freshness and disclosure controls122 from applicable configuration and the brief, not a fixed score or age proxy.123 - A required missing tool/control remains blocked; neutral mode needs no logo.1241252. **Read brief + extract structure:**126 - Goal / key message (one sentence)127 - Audience profile128 - All substantive sections, claims/evidence, tables, citations and material limitations129 - Opening context and closing decision/CTA where appropriate130 - Retain full source separately from the concise slide view131132 **Standalone count selection:** Explicit `--slide-count` takes precedence.133 Honor brief/duration constraints before the retained 20-30 slides based on134 duration starting range; material, audience and pacing can justify an135 explained adjustment. This is an advisory fallback, not a mandatory gate,136 not a Word/web section quota, and not the outline stage's 8-15/default-12 hint.137 Surface conflicting explicit requirements without deleting source content.1381392b. **Design DNA slide pass (ADR-0017 — retrieval before slide design).** Query the slide140 decision engine so the arc is grounded in the corpus, not invented:141 ```bash142 dna="${LINTEL_SOURCE_ROOT:?select the trusted source}/skills/design-dna"143 python3 "$dna/scripts/search.py" "<deck goal / pitch type>" --slide strategy -n 1 # narrative arc + sparkline-beats144 # then per slide, by the slide's emotion (trust|urgency|confidence…) and goal (hook|proof|cta…):145 python3 "$dna/scripts/search.py" "<emotion>" --slide color-logic -n 1 # background/text/accent + full-bleed146 python3 "$dna/scripts/search.py" "<goal>" --slide layout-logic -n 1 # layout pattern + break-pattern147 python3 "$dna/scripts/search.py" "<slide-type>" --slide copy -n 1 # headline formula148 ```149 Feed the strategy's `sparkline_beats` + `emotion_arc` to PPTNarrativeArchitect as the arc spine.150 python3 absent → Read `design-dna/data/slides/*.csv` directly (controlled vocabulary in design-dna SKILL.md).1511523. **Use the accepted read-only `PPTNarrativeArchitect` method** to design the arc;153 delegate only through a real available host operation, otherwise label the154 builder's own pass as self-review:155 - Slide-by-slide content goals — seeded by the slide-strategy `sparkline_beats`156 - Mode tags from the configured corpus when applicable; neutral is valid157 - Layout suggestions per slide — from `--slide layout-logic` (pattern + break-pattern at 1/3, 2/3)158 - Asset suggestions (from the active pack's asset library, if one is configured)1591604. **Generate editable slides through the selected native API or pptxgenjs:**161 - Apply an explicit compatible template or the requested neutral blank design162 - Add slides per architect's arc163 - Compose readable visible text without changing the scope of source claims164 - Insert assets from the active pack's asset library where matched165 - Write complete supporting paragraphs, citations, tables and the claim ledger166 into actual speaker notes, an appendix or delivered linked long-form content.167 Always retain the full source. Put a material limitation on the visible slide168 whenever omitting it would misrepresent a visible claim.169 - Keep source §N IDs and record continuation/appendix locations. A slide-count170 target cannot authorize losing facts, shrinking illegibly or clipping text.1711724b. **Critique narration using the accepted read-only `SlideNarrationCritic`173method** when a talk track exists. Supply the exact deck/notes, intended audience,174duration and current profile reference through the available host delegation175operation. Check pacing and recovery lines without deleting supporting evidence.176Keep speaker talk time separate from appendix/reference material. A claimed177words-per-minute target is not an observed rehearsal.1781795. **Inspect the saved artifact, then apply the familiar control categories:**180181 Run the shipped read-only [`check_pptx.py` source-retention procedure](../../../skills/generate-ppt/references/retention-check.md)182 on the actual deck and explicit source paragraphs/cells. It checks bounded183 ZIP/XML and real note relationships; keep missing detail and malformed/aliased184 notes as findings. This does not supply application or rendered-layout evidence.185186 First reopen through the actual available application/API, make a small scoped187 edit and read it back. Compare full source to slides plus actual saved notes.188 Render every slide using the available renderer and inspect wrapping, overflow,189 overlap, table legibility and assets. Check notes separately; slides alone cannot190 prove note retention. Fix layout by reflow/splitting, not by discarding facts.191192 **Voice:** apply only the verified corpus's requirements and applicability.193 Neutral clarity/pacing advice does not become a hard vocabulary or score gate.194195 **Brand:** check configured template/tokens and explicitly authorized local196 assets. No invented logo, personal library scan or claim that a blank deck is197 company-branded. Missing mandatory brand data cannot silently fall back.198 Use the named `brand-source` and `brand-freshness`199 [selected asset evidence procedures](../../../skills/design-dna/references/design-contract.md#selected-asset-evidence).200 Record actual source/version and refresh requirements as P05 `check` controls;201 unknown currency stays unverified. `--ignore-stale-brand` records only a202 permitted advisory exception, not a passed mandatory freshness check.203204 **Honest limitations:** check each material claim against its evidence,205 assumptions and failure boundary, including claims on pitch/workshop slides.206 Disclosure requirements follow the brief/policy; counts cannot prove honesty.207 Record a grounded N/A only for a genuinely inapplicable requirement.208209 **Provenance:** bind the exact brief, source content, notes, template/config,210 output and live P07 reference using existing P05 snapshot/control evidence.211 File metadata or a made-up provenance ID cannot replace those observations.2122136. **Report the actual result at the explicit owned output path.**214 Mandatory failure/error/unverified blocks the affected acceptance or215 distribution regardless of scores. Retain useful editable output and its216 source while naming any missing renderer or review. Do not use an implicit217 personal draft directory. An API edit/readback is not every-client acceptance.218219## Reusable patterns220221Follow the [reusable pattern consumer contract](../../../skills/pattern/references/consumer-contract.md).222Brief mode resolves for the brief's context; from-pipeline mode verifies the run's223`pattern_context` attachment. Both receive the same clauses before the slide structure is224chosen: required sections become slides or sections, and the clauses go to the existing225narrative review and the QA report. A helper fixture proves only this integration, not a real226generated deck.227228## Report format229230Record exact source/output paths and hashes; actual profile/template; audience and231arc; slide-to-section/notes mapping; writer and native provider/instance/actions;232edit/readback result; rendered slides inspected and specific renderer limits;233retained claims, citations, table data and material limitations; configured234control outcomes; and remaining acceptance/review gates. Keep documentary checks,235package extraction, native actions and render evidence separate.236237## Compliance integration238239- Configured applicable controls keep their mandatory/advisory semantics.240- No customer data, secrets, macros, external upload or automatic asset download241 in synthetic validation. Use explicit authorized local assets.242- Required independent review and delivery controls remain separate from creation.243244## Voice tier note245246`voice: mixed`. Skill itself is engineering-internal; the SLIDE CONTENT may use a customer-facing voice tier (gated by Gate 1).247248## Failure modes249250- **Writer error** — preserve source and the owned draft, record the actual error; do not emit a passing gate.251- **Required template missing** — block; offer neutral defaults only when applicable policy permits.252- **Voice gate fails after 2 regen attempts** — keep draft, surface specific slide failures with fix recommendations253- **Required asset unavailable** — report its exact gap; do not invent a successful asset inspection.254- **Renderer absent** — retain editable deck/source but keep rendered-layout acceptance unverified.255- **Customer-data in brief** — BLOCK before generation. Sanitize first.256257## Examples258259**Standard pitch deck:**260```261> /generate-ppt --brief docs/engagement/customer-A-pitch-brief.md --template pitch-deck --audience "public-sector CIO"262[Resolve the explicit template, preserve evidence, and inspect the actual artifact.]263```264265**Workshop deck with defaults:**266```267> /generate-ppt --brief synthetic-brief.md --template workshop --use-defaults --voice internal --out workshop.pptx268[An explicitly blank native deck is valid neutral output; configured requirements still apply.]269```270271**Stale brand override:**272```273> /generate-ppt --brief archive-pitch.md --template pitch-deck --ignore-stale-brand "archival deliverable, brand version pinned"274[Records an advisory exception; does not waive required brand policy.]275```276277## See also278279- [Native PowerPoint procedure](../../../skills/generate-ppt/references/native-powerpoint.md)280- [Content fidelity and evidence](../../../skills/generate-write/references/fidelity-and-evidence.md)281- `PPTNarrativeArchitect` agent — slide arc design282- `SlideNarrationCritic` agent — slide narration / talk-track critique (post-generation)283- The active pack's compliance gates — Gate 1284- `/generate-word`, `/generate-web` — sibling doc-gen skills285- [Brand source and freshness evidence](../../../skills/design-dna/references/design-contract.md#selected-asset-evidence) — scoped inspection, not a hook