Research shortform

Short-Form Research — Orchestrator

Pipeline skill — produces the per-platform best-practice catalog that brief-shortform consumes per asset.

Core Question: "What's working right now on short-form for our topic and market — and which patterns should the next 30 days of briefs bet on?"

[Read references/playbook.md [PLAYBOOK] for why this skill exists, methodology, principles, and when NOT to use.]

---

Critical Gates — Read First

Non-negotiable constraints before dispatching any agent:

1. No fabricated data. Every claim, number, and pattern must trace to a source URL, video ID, or cited platform doc. Orphan claims fail critic rubric #1.
2. Single market per artifact. Multi-market campaigns re-run research per market — never mix VN and US findings in one artifact. Cultural patterns are not averageable.
3. Hard cap on platforms. Default 3 (TikTok + Reels + Shorts). X video and LinkedIn video are explicit opt-in via --all or --platforms. Maximum ever is 5. Cost discipline.
4. Sample-size honesty. Every per-platform section declares OK (n≥8), LOW_SAMPLE (n=3-7), or INSUFFICIENT_DATA (n<3). LOW_SAMPLE flags carry through to brief skill warnings. INSUFFICIENT_DATA means no pattern claims at all — only observed examples.
5. Two freshness windows. Trend signals refresh every 14d, warn at 30d. Platform mechanics refresh every 90d, warn at 180d. Frontmatter records mechanics_sources_verified[] — the actual doc URLs and their last-updated dates, not just the run timestamp.

Quality Gate

Critic agent verifies before delivery (all five PASS required, max 2 rewrite cycles):

• Every numerical claim and named pattern has a source URL or video ID
• Every per-platform section declares OK / LOW_SAMPLE / INSUFFICIENT_DATA per the N≥8 / 3-7 / <3 rule
• Every recommendation is platform-specific (could not be moved to another platform's section unchanged)
• Every cited mechanic links to a source doc with a verified last_updated date inside the 180d warn window
• Audience Fit section either references ICP or explicitly declares "no ICP — using cold-start hint" with the hint text included

---

Before Starting

Apply the before-starting-check [PLAYBOOK]:

0. Mode resolution per references/_shared/mode-resolver.md [PROCEDURE]. Skill is budget: deep; --fast collapses to single-pass scout + synthesis with critic skipped, but Critical Gates above STILL enforced (safety supersedes --fast). Cold Start (Pre-Dispatch) still fires under --fast if topic/market are missing. Session execution profile (single-vs-multi): inherit per references/_shared/execution-policy.md.
1. Read implementation-roadmap/canonical-paths.md if present — verify output path matches canonical inventory.
2. Read .forsvn/index/manifest.json — check for prior short-form-research artifacts under (topic, market) for warm-start eligibility.
3. Run Pre-Dispatch per references/procedures/pre-dispatch.md [PROCEDURE] — needed dimensions, read order, Cold/Warm prompts, write-back map all there.

Artifact Contract

• Path: docs/forsvn/artifacts/research-research-shortform-<YYYY-MM-DD>-<slug>.md (flat v2 grammar; one artifact per topic+market+platform-set)
• Lifecycle: pipeline (per research-skills/CLAUDE.md; canonical-paths.md notes this is borderline-canonical — consumed cross-stack — but pipeline classification preserved verbatim for backwards-compat; refresh trigger handled by freshness windows, not manifest archival)
• Frontmatter fields: skill, type, status, date, stack (=research), review_surface (=md — pipeline defaults to decision_state: not_required), topic, market, platforms_analyzed, platform_mechanics_date, mechanics_sources_verified[], trend_signals_date, sample_size_per_platform, icp_referenced (full schema in Output Artifact Structure below; cross-stack v2 contract in references/_shared/artifact-contract-template.md)
• Required sections (in order): TL;DR · Audience Fit · Per-Platform Findings · Cross-Platform Comparison · Trending Audio (conditional) · Recommendations for short-form-brief · Open Risks & Caveats · What This Research Doesn't Cover
• Consumed by: brief-shortform (marketing-skills) per-asset, reads §6 Recommendations + frontmatter sample-size flags; evaluate-shortform (research-skills) cycle-N scorer, reads frontmatter + §3 Per-Platform Findings to score published posts against the catalog
• Cross-stack contract: schema changes require atomic update of BOTH consumers — never silently drift the frontmatter or section order (per anti-patterns.md row "Cross-stack contract drift")

---

Agent Manifest

Agent	Layer	File	Focus
Platform Scout	1 (parallel — N×)	agents/platform-scout-agent.md	Per-platform top performers via WebSearch + WebFetch; URLs + metrics + opening 1-3s + audio + caption + CTA
Audience Fit Agent	1 (parallel)	agents/audience-fit-agent.md	ICP / product-context / cold-start hint → register, language polish, sensitivity flags
Pattern Extractor	2 (sequential)	agents/pattern-extractor-agent.md	Recurring hook archetypes per platform + sample-size flags
Audio Trend Agent	2 (sequential, conditional)	agents/audio-trend-agent.md	Only if TikTok/Reels in scope; trending sounds + usage counts + decay risk
Synthesis Agent	2 (sequential)	agents/synthesis-agent.md	Writes artifact: TL;DR, per-platform, comparison, recommendations, risks
Critic Agent	2 (final)	agents/critic-agent.md	Five-rubric quality gate; routes rewrites; max 2 cycles

Routing + Dispatch

Single route — full Layer 1 + Layer 2 sequence runs every time (per-platform analysis IS the value):

1. Pre-Dispatch (warm-start scan + cold-start if needed) — per procedures/pre-dispatch.md
2. LAYER 1 IN PARALLEL: platform-scout × N + audience-fit-agent
3. LAYER 2 SEQUENTIAL: pattern-extractor → audio-trend (conditional) → synthesis → critic
4. Critic FAIL → re-dispatch named agent(s) (max 2 cycles); after 2, ship done_with_concerns
5. Deliver artifact

Mechanics (how to spawn agents, parallel/sequential tables, critic routing rules, single-agent fallback, chain position, re-run triggers, skill deference) live in references/procedures/dispatch-mechanics.md [PROCEDURE]. Load at Layer 1 dispatch entry.

---

Anti-Patterns

Critic-load reference: references/anti-patterns.md [ANTI-PATTERN]. Re-read before any output ships — covers orphan claims, multi-market mixing, sample-size dishonesty, stale mechanics, generic recommendations, critic-loop overrun, cross-stack contract drift, and 8 more failure modes.

Durable Rules (protected)

<!-- SLOW_UPDATE_START --> <!-- No pinned rules yet. Populate via the slow-update workflow (see references/slow-update-fence.md). Each pinned rule must (a) be procedural not instance-specific, (b) be earned from a regression or critic-flagged failure, (c) cite the artifact / decision record that justified pinning. --> <!-- SLOW_UPDATE_END -->

Completion Status

Skill returns one of:

• DONE — all 5 critic rubrics PASS within ≤2 cycles. All requested platforms returned ≥3 entries.
• DONE_WITH_CONCERNS — critic loop cap reached; remaining failures are surfaceable as warnings (LOW_SAMPLE on 1+ platforms, one stale source doc beyond warn window). Concerns pinned at top of artifact.
• BLOCKED — WebSearch / WebFetch blocked or rate-limited; ICP read failed. Requires user action — state what's needed.
• NEEDS_CONTEXT — cold-start abandoned; or audience_hint empty AND no ICP. Defer to research-icp.

References

• references/playbook.md [PLAYBOOK] — why this skill exists, methodology, principles, history, when NOT to use
• references/_shared/before-starting-check.md [PLAYBOOK] — pre-Pre-Dispatch read pattern (canonical at references/, synced)
• references/_shared/mode-resolver.md [PROCEDURE] — --fast behavior contract
• references/_shared/pre-dispatch-protocol.md [PROCEDURE] — canonical Pre-Dispatch spec
• references/procedures/pre-dispatch.md [PROCEDURE] — this skill's Cold + Warm Start prompts + write-back map
• references/procedures/dispatch-mechanics.md [PROCEDURE] — Layer 1/2 spawn mechanics, critic routing, single-agent fallback, chain position, skill deference
• references/anti-patterns.md [ANTI-PATTERN] — failure modes
• references/format-conventions.md [PROCEDURE] — date format, URL handling, citation, sample-size flag, per-platform ordering
• references/scoring-rubrics.md — pattern-extractor + critic rubric definitions
• references/_shared/confidence-labeling.md [PROCEDURE] — canonical H/M/L confidence label + L-resolution rule; the claim-level certainty tag layered on the sample-size flags
• references/scout-protocol.md — per-platform sourcing protocol
• references/platforms/ — per-platform research playbooks (tiktok, instagram-reels, youtube-shorts, twitter-video, linkedin-video, _comparison)
• research-skills/CLAUDE.md §"Pre-Dispatch Protocol" + §"Complexity Routing" + §"Multi-Agent Skills" — stack-level conventions this skill inherits