Generate

---
description: "Build a source-backed AI industry briefing from official vendor publications, configured RSS feeds, GitHub releases, reputable secondary reporting, and user-supplied URLs. Use when: 'ai briefing', 'ai news', 'what's new in AI', 'catch me up on AI', 'prep for AI meeting', 'AI roundup', or 'generate AI slides'."
argument-hint: "[retro|search] [--profile <name>] [--since <1d|7d|30d>] [--format markdown|slides|html] [--yes]"
user-invocable: true
disable-model-invocation: false
---

## Variables

Arguments: `$ARGUMENTS`
Configured active profile: `${user_config.active_profile}`

## Purpose

Collect, deduplicate, rank, and present current AI-industry developments without depending
on organization-specific configuration. Markdown is the default output. HTML and PPTX are
explicit, optional outputs backed by the bundled deterministic build pipeline.

## Source and access policy

Use only interfaces that authorize automated access:

1. Official vendor blogs, documentation, changelogs, and release notes.
2. GitHub release pages and repository APIs.
3. RSS/Atom feeds listed by the selected profile.
4. Reputable secondary reporting, corroborated by a primary source when practical.
5. User-supplied public URLs and context.

Automated X/Twitter collection is disabled. Do not crawl, scrape, scroll, navigate, fetch,
or extract X pages through browser automation, DOM scripts, unofficial endpoints, or Grok.
Do not refresh following graphs or profile metadata. X's current terms expressly prohibit
crawling or scraping without prior written consent: <https://x.com/en/tos>.

If the user supplies an X URL, preserve it as user-provided citation metadata but do not
programmatically retrieve it. Ask for the relevant text when it is necessary, and seek a
non-X primary source for corroboration. This plugin does not configure a paid X API.

Playwright is allowed only for deterministic rendering and inspection of locally generated
HTML/PDF artifacts. It is not a collection provider.

## Arguments

| Flag | Default | Effect |
|---|---|---|
| `--profile <name>` | Configured `active_profile`, then `default` | Select a named profile for this invocation. |
| `--since <timeframe>` | 14d | Accept `1d`, `3d`, `7d`, `14d`, `30d`, or an ISO date. |
| `--providers <list>` | All | Limit collection to comma-separated provider buckets. |
| `--extras` / `--no-extras` | On | Include or omit robotics, science, and novel applications. |
| `--format <type>` | `markdown` | Emit markdown, HTML, or PPTX. Non-markdown formats require prior `/ai-briefing:setup apply install-build-deps`. |
| `--yes` / `-y` | Off | Skip the pre-execution confirmation gate. Required for headless runs. |

Actions:

- Default: collect, deduplicate, rank, cite, and emit a briefing.
- `retro --meeting <N>`: record acted/noted/skipped feedback for an archived briefing.
- `search "<query>"`: search archived briefing markdown and return cited matches.

## Profile resolution

Files at `.claude/ai-briefing/` form the default profile. Each
`.claude/ai-briefing/<name>/` directory is a named profile. Selecting a named
profile is profile selection, not resolution of a `*.local.*` cascade layer.
Resolve the active profile in this order:

1. `--profile <name>` for this invocation.
2. The rendered `${user_config.active_profile}` value when it is non-empty.
3. `default`.

Require a 1-63 character lowercase-kebab profile name and reject reserved Windows device
names before constructing any profile path.

Resolve that value in the skill before launching tools. Explicitly pass the result to every
build subprocess; do not assume plugin configuration is inherited by an arbitrary shell:

```bash
CLAUDE_PLUGIN_DATA="${CLAUDE_PLUGIN_DATA}" AI_BRIEFING_PROFILE="$PROFILE" node "${CLAUDE_PLUGIN_DATA}/runtime/build/run.js"
```

Set `CLAUDE_PLUGIN_DATA` on the launch the same way. The Bash tool's environment does not carry
this plugin's value, and another plugin's SessionStart hook can export its own data directory
under that name, so the build uses an inherited value only when it names ai-briefing. Basis: the
plugins reference, <https://code.claude.com/docs/en/plugins-reference#where-each-variable-resolves>,
verified 2026-10-03; recheck when that table lists the Bash tool among the processes that receive
the variables.

Keep the selection per-invocation: set it on each launched process as above, never as a global
export the consumer adds to their shell profile. A global export pins one profile for every
session and every other tool on that machine, including runs that asked for a different one.

A profile may contain these team-tracked files. They are not a `*.local.*`
cascade overlay:

| Artifact | Purpose |
|---|---|
| `sources.md` | Approved RSS/Atom feeds, official release pages, GitHub repositories, and user-supplied URLs. |
| `audience.md` | Optional stack/audience lens for impact annotations. |
| `brand.json` | Optional declarative deck brand file. |

## Default run

1. Parse and validate arguments. Reject unknown formats and malformed dates before any
   collection. Resolve the selected profile and read `sources.md` when present.
2. Unless `--yes` is present, show the timeframe, provider scope, source classes, selected
   profile, and requested output format. Ask for confirmation before outbound collection.
3. Search official vendor publications and GitHub releases first. Read configured feeds and
   permitted user-supplied non-X URLs. Use reputable secondary reporting to fill gaps and
   corroborate claims. Record source URL, publisher, publication date, and retrieval date.
   Set an explicit timeout on every outbound request, and keep partial failures visible
   instead of dropping the source silently.
4. For every item, require at least one working source URL. Prefer the primary announcement;
   label claims that remain secondary-only. Exclude content outside the requested window.
5. Deduplicate by canonical URL and normalized event identity. Merge corroborating sources
   into one item rather than repeating the same announcement.
6. Categorize into provider buckets and rank by practical impact, availability, novelty,
   source confidence, and relevance to `audience.md`. Never invent details or silently turn
   rumors into facts.
7. Write the current markdown briefing. Every requested provider bucket appears in it,
   including one the window produced no items for; say that the bucket is empty rather than
   omitting it silently. Update the seen-item registry under the selected profile's
   `${CLAUDE_PLUGIN_DATA}` state directory only after the markdown is emitted successfully.
   Keep tracked profile configuration in the project; never write curated configuration into
   plugin data. Re-running the same window is idempotent: merge by the same normalized event
   identity step 5 deduplicates on
   and do not emit duplicate items.
8. For `--format html` or `--format slides`, require the optional build tree installed by
   `/ai-briefing:setup apply install-build-deps`. Run the staged build pipeline against the emitted
   markdown with `CLAUDE_PLUGIN_DATA="${CLAUDE_PLUGIN_DATA}"` and `AI_BRIEFING_PROFILE="$PROFILE"`
   set on the launched process, as in the launch line above. Playwright may
   open only generated local HTML for PDF rendering and layout validation. Surface missing
   prerequisites with the exact setup command; do not silently fall back to another browser
   provider.
9. Lead the report with anything waiting on the user (a missing build prerequisite with its
   setup command, the text of a user-supplied X URL the run needs), then the collected,
   excluded, deduplicated, and secondary-only counts plus the output path.
   Make warnings visible, including unreachable sources and missing optional build tooling.

## Output contract

The markdown briefing contains:

- Run metadata: profile, window, retrieval time, and source classes used.
- Provider sections with HIGH, MEDIUM, and LOW tiers.
- Item title, concise evidence-backed summary, practical impact, publication date, and
  source links.
- A visible caveats section for inaccessible sources, secondary-only claims, and gaps.
- An optional EXTRAS section when enabled.

Machine-local state and generated artifacts live under
`${CLAUDE_PLUGIN_DATA}/<profile>/`. Tracked source, audience, and brand configuration lives
under `.claude/ai-briefing/[<profile>/]` in the consumer repository.

## Quality gates

- Every factual item has a source URL and date.
- Primary sources are preferred and clearly distinguished from secondary reporting.
- No automated X access occurs, including through browser tools or indirect DOM scripts.
- The same event appears once, with corroborating links attached.
- Missing dependencies and partial collection are visible; no black-box degradation.
- HTML/PDF browser use is local-render-only.

## Reference index. Load on demand

| File | Load when |
|---|---|
| `reference/audience-defaults.md` | Step 6 ranking, when no profile `audience.md` supplies the lens or a profile only partly overrides it. |
| `reference/build-pipeline.md` | Step 8, running the build: commands, the `slides-data.js` schema, and its prerequisite checks. |
| `reference/slide-generation.md` | Step 8 for `--format slides`, deciding slide order, split rules, and brand tokens. |