aget-create-briefing

name	aget-create-briefing
description	Generate standalone narrative documents from session artifacts, optimized for a target consumption medium (NotebookLM, slides, memo, share-draft). Use when the user says "create briefing", "create a briefing for NotebookLM", "make this listenable", "summarize session for [medium]", "prepare a listen", "brief me on [topic]", or "create a summary I can share".

/aget-create-briefing

Generate standalone narrative documents from session artifacts, optimized for a target consumption medium. Bridges internal KB formats to external-consumption artifacts.

Purpose

AGET sessions produce knowledge in internal formats (L-docs, session records, ontology entries, project plans) optimized for machine-readability and cross-reference integrity. These formats are unsuitable for external consumption — whether by the principal reviewing via audio (NotebookLM), preparing a presentation, or sharing a summary.

This skill bridges that gap: select source artifacts, adapt format to target medium, strip governance ceremony, and apply audience-appropriate voice register.

Evidence: Two independent demand signals (L758 cross-agent convergence). CCB NotebookLM briefing (~3000 words, 2026-04-02). Framework-AGET state management briefing (2026-03-30). L384 documents the KB-to-narrative-to-NotebookLM pipeline as established practice.

Input

$ARGUMENTS

Parameters (all optional — defaults apply):

Parameter	Values	Default	Purpose
--source	Session record path, L-doc ID(s), topic string, or latest-session	latest-session	What to synthesize
--medium	notebooklm, slides, memo, share-draft	notebooklm	Target consumption format
--audience	self, principal, external	principal	Who consumes this
--register	cognitive, formal, conversational	Per agent config or conversational	Voice register

Examples:

• /aget-create-briefing — Briefing from latest session, NotebookLM format, principal audience
• /aget-create-briefing --source L758 --medium memo — Memo from specific L-doc
• /aget-create-briefing --source sessions/SESSION_2026-04-04_*.md --medium slides — Slides from specific session
• /aget-create-briefing --medium share-draft --audience external — Public-safe draft from latest session

Execution

Step 1: Parse Parameters

Parse $ARGUMENTS for --source, --medium, --audience, --register. Apply defaults for any omitted parameters.

If .aget/config.json contains a briefing: block, use agent-specific defaults from it. Otherwise use the universal defaults from the parameter table above.

Step 2: Source Resolution

Resolve the --source to actual KB artifacts:

Source Type	Resolution
latest-session (default)	Find most recent sessions/SESSION_*.md by date
Session path	Read the specified session record directly
L-doc ID(s)	Read .aget/evolution/L{id}_*.md for each ID. Multiple IDs: comma-separated (e.g., L758,L771)
Topic string	Run python3 scripts/study_topic.py --topic "{topic}" --quiet and read the top 5 results

Verify: Confirm at least 1 source artifact was resolved. If 0, report error and stop.

Step 3: Content Extraction

Read all resolved source artifacts. Extract:

1. Claims: Key findings, decisions, patterns discovered
2. Evidence: Data points, exemplars, citations supporting claims
3. Open questions: Unresolved issues, decision points pending
4. Context: Why the work was done, what problem it addresses

Strip: Remove all governance ceremony — gate references (G0, G1...), spec IDs (CAP-xxx, R-xxx), EARS patterns (SHALL/WHEN/WHERE), conformance scores, V-test results, deliverable checklists, velocity metrics. Preserve the substance these structures contained, not the structures themselves.

Step 4: Medium Adaptation

Apply format rules based on --medium:

notebooklm (audio-optimized):

• Plain prose only. No tables, no YAML, no code blocks, no bullet lists longer than 3 items (NotebookLM struggles with structured data)
• Conversational structure with implicit question-answer flow
• Use transitions that work when spoken: "The key finding here is..." / "What makes this interesting is..." / "The question this raises is..."
• Target: 1500-3000 words (10-20 min listen)
• Sections as narrative flow, not headers (minimize headers — NotebookLM reads them awkwardly)

slides (visual-optimized):

• One assertion per section, bullet-heavy, minimal prose
• Each section: title (the assertion) + 2-4 supporting bullets + optional speaker note
• Target: 15-25 sections
• End with "Key Takeaways" section (3-5 bullets)

memo (text-optimized):

• Executive summary (3-5 bullets) at top
• Detail sections with headers, concise paragraphs
• Formal register regardless of --register parameter (memos are always formal)
• Target: 800-1500 words

share-draft (public-safe):

• Apply VOICE.md anti-patterns if available, otherwise use neutral professional tone
• MANDATORY: Run classification safety check (Step 5) before writing output
• Include [DRAFT — requires principal review before external use] watermark at top
• Target: 500-2000 words depending on source density

Step 5: Classification Safety Check (share-draft only)

MANDATORY for audience=external or medium=share-draft. Skip for self/principal audiences.

Scan the draft output for these prohibited patterns:

Pattern	Example	Action
Private agent names	private-*-aget, private-*-AGET	Remove or replace with generic description
Private repo references	gmelli/*	Remove or replace with "internal repository"
Fleet size disclosures	"32 agents", "40 agents in fleet"	Remove or replace with "multiple agents"
Internal project IDs	FLEET-*-###, REL-042	Remove
Session references	SESSION_2026-*	Remove or replace with "a recent session"
Internal L-doc references	L758, L771	Remove or replace with natural language description

If any pattern found: Fix it in the draft, then re-scan. Report what was sanitized.

Step 6: Write Output

Write the briefing to docs/briefings/ with the naming convention:

Medium	Filename
notebooklm	BRIEFING_YYYY-MM-DD_{topic}.md
slides	BRIEFING_YYYY-MM-DD_{topic}_slides.md
memo	BRIEFING_YYYY-MM-DD_{topic}_memo.md
share-draft	BRIEFING_YYYY-MM-DD_{topic}_draft.md

Where {topic} is a snake_case summary of the source content (max 50 chars).

Step 7: Report

Display completion summary:

=== Briefing Created ===
File: docs/briefings/BRIEFING_YYYY-MM-DD_{topic}.md
Medium: {medium}
Audience: {audience}
Register: {register}
Sources: {count} artifacts ({list})
Words: {word_count} (~{estimated_minutes} min {read/listen})
Classification: {PASS | N/A (internal audience)}

Constraints

• C1: MUST NOT modify any source KB artifacts. This skill is read-only on the KB. Output goes only to docs/briefings/.
• C2: MUST NOT produce output for audience=external without completing Step 5 (Classification Safety Check). Zero tolerance for private pattern leaks.
• C3: MUST NOT include governance ceremony in principal/external output — no CAP-xxx, R-xxx, EARS patterns, V-test results, gate references, conformance scores.
• C4: MUST NOT fabricate claims. Every substantive assertion in the briefing must trace to a source artifact. If evidence is tentative, say so. If uncertain, omit rather than guess.
• C5: MUST validate that at least 1 source artifact resolves before proceeding to Step 3. Report error on 0 sources.
• C6: MUST use docs/briefings/ as output directory. Create the directory if it does not exist.

Error Handling

Error	Cause	Resolution
No source artifacts found	--source resolved to 0 files	Report: "No artifacts found for source '{source}'. Try a different source or check the path."
Session file not found	latest-session but no sessions/SESSION_*.md exists	Report: "No session files found in sessions/. Provide an explicit --source."
L-doc ID not found	Specified L-doc doesn't exist in .aget/evolution/	Report: "L{id} not found. Check .aget/evolution/ for available L-docs."
Classification check failed	Private patterns in share-draft after 2 sanitization passes	Report: "Classification safety check failed after 2 passes. Manual review required." Refuse to write output.
Unknown medium	--medium value not in {notebooklm, slides, memo, share-draft}	Report: "Unknown medium '{value}'. Use: notebooklm, slides, memo, share-draft."

Related Skills

• /aget-study-topic — Research a topic across KB (often the precursor to creating a briefing)
• /aget-record-observation — Capture findings during sessions (produces the artifacts briefings synthesize)
• /aget-record-lesson — Record L-docs (a common briefing source type)
• /aget-wake-up — Session initialization (briefings often cover session output)

Traceability

Link	Reference
Proposal	SP-005 (planning/skill-proposals/PROPOSAL_aget-create-briefing.md)
Requirements	REQ-BRF v0.1.0 (docs/drafts/REQ-BRF_briefing_creation.md)
Tracking	gmelli/aget-aget#810
L-docs	L384 (Spec Narrative Pattern), L733 (Voice as Composition), L745 (Multi-Modal Effectiveness), L758 (Cross-Agent Convergence), L771 (Proposal Discovery Gap)
Cross-agent	CCB P-005 (demand signal + parameter design + working exemplar)

---

aget-create-briefing v1.0.0 Category: Creation Archetype: Universal (all agents, medium/audience parameters adapt per agent identity) Template: templates/skill/SKILL.template.md