Triage v0 Spec — `/triage-inbox` (lite)

Status: APPROVED 2026-07-17 (scope v0-lite, spec, bolding deferral — task file Ideas-Workflow-triage-v0). Skill built: ~/ai-config/skills/triage-inbox/SKILL.md (ai-config c4ebcd9). Shared deliverable with KB-OS M4.

Scope decision (recommended): v0-lite — Clippings only

v0 triages Core/Misc/Clippings/ only. Dept Inboxes are out of scope until v1.

Rationale:

The immediate consumer is P3: the 93-clipping backlog. Clippings are the unvalidated firehose; dept Inbox bullets were already human-routed at capture time — different, smaller problem.
18 dept Inboxes with several long ones make full M4 scope meaningfully bigger; v0-lite validates the mechanism on real volume first.
The skill is named and arg-shaped for the full scope from day one (below), so v1 extends it — no rename, no rework.

Skill name + invocation

/triage-inbox [source] — named for what it does (triage intake sources), matching the KB-OS M4 design note.

/triage-inbox or /triage-inbox clippings — v0: triage Core/Misc/Clippings/
/triage-inbox <dept> / /triage-inbox all — v1+ (rejects with “not yet implemented” in v0)
Optional: --batch N (default 12) — number of un-triaged clips per pass

Per-item processing (keep portable)

P3 may run on a free/local model (Gemini API, local). The per-item prompt must stay simple: input = clip frontmatter + body; output = 4 fields. No vault-wide context beyond two small reference tables loaded once per batch (save criteria from Capture-Routing; Biggest Rocks table from Core/DASHBOARD.md).

Per clip, produce:

Field	Spec
TLDR	1–2 sentences, what it says + why it might matter to Talbot’s businesses
Rating 0–10	Rubric below
Route	One of: `KEEP` / `INBOX → <dept>` / `TASK → <dept>` / `DISCARD`
Why	One line justifying route + rating

Flag ⭐ ROCK on any clip whose idea directly advances a current Biggest Rocks row — these get elevated to the top of the batch report.

Rating rubric (anchored to save criteria)

8–10 — directly actionable for a current Biggest Rock or active task; or unique un-regenerable reference
5–7 — probable future reference value, clearly relevant to a dept’s scope
2–4 — generic/regenerable by AI on demand; weak business fit
0–1 — fails save criteria outright (pure news, duplicate, clickbait with no substance)

Route definitions (lands in the existing Capture-Routing pipeline — never a parallel one)

KEEP — stays in Clippings/ as reference; frontmatter marked triaged
INBOX → — one bullet appended to the owning dept’s Inbox.md (Capture Stage 1), linking the clip; clip stays as the source
TASK → — promote-worthy now: bullet appended to dept Inbox.md prefixed [PROMOTE] — Talbot runs the Templater promote wizard himself (Stage 2 stays human)
DISCARD — fails save criteria: moved aside to Core/Misc/Clippings/archive/ (never deleted — move-aside rule)

dept: frontmatter is a hint, not truth — the skill may propose a different dept with justification.

Batch flow (human-gated — skill never routes unilaterally)

Scan — find un-triaged clips (no triaged: frontmatter), take batch of N
Report — one markdown table: ⭐ | # | Clip | Rating | Route | TLDR | Why, ROCK-flagged rows first, then by rating desc
Approve — Talbot replies in chat: approve all, or overrides by row (3 → DISCARD, 5 → INBOX Mktg), or skip 4
Execute — only after approval: append Inbox bullets, move DISCARDs to archive/, write back frontmatter to every processed clip:
```
triaged: 2026-07-17
rating: 7
routed-to: "INBOX → MBR/Mktg"   # or KEEP / TASK → ... / DISCARD
```
Summarize — counts per route + list of Inbox files touched

Frontmatter write-back is what makes P3 batch sessions resumable — re-running the skill skips triaged clips automatically.

Out of scope for v0 (deferred)

Dept Inbox triage (v1 — the <dept>/all args)
Key-phrase bolding in kept clips (optional flag later; adds body-rewrite risk for marginal value — recommend deferring)
Auto-promote to TaskNote (Templater wizard is interactive; Stage 2 stays human)
Dept inference in /task-start//task-from-chat (adjacent KB-OS M4 item, explicitly not this task)

Build hand-off (AI dept)

Home: ~/ai-config/skills/triage-inbox/SKILL.md (+ Claude command wrapper per deploy convention)
Build model: Sonnet, after this spec is approved
Constraints: clean ai-config tree before commit (post-commit deploys working tree); Skills Quality Standard (review → optimize → test); update Core/AI/Skills/DASHBOARD.md
Acceptance test: Coconote clip end-to-end (clip → triage table → approve → frontmatter written + Inbox bullet landed). Spec-level dry run: appendix below.

Appendix — dry-run on acceptance clip (spec validation, no files modified)

Input: 🥥 The Hidden Secret Behind Coconote….md (dept: AI hint, tags: clippings)

⭐	#	Clip	Rating	Route	TLDR	Why
	1	🥥 Coconote $6.7M ARR	6	INBOX → MBR/Mktg	AI note-taking app hit $6.7M ARR / Quizlet acquisition in 18 months with zero paid ads — growth driven by organic social (1B+ views). Playbook relevant to MBR viral-growth marketing.	Value is the zero-ad organic-growth playbook, not the AI tooling — `dept: AI` hint overridden; regenerable story but tactic list has reference value → keep + route bullet

Proposed Inbox bullet (MBR/Mktg/Inbox.md): - [clip](/core/misc/clippings/🥥-the-hidden-secret-behind-coconote-how-an-ai-tool-hit-$6.7m-arr-in-just-18-months/) rated 6: Coconote growth playbook — $6.7M ARR in 18 mo, zero paid ads, 1B+ organic social views; triage 2026-07-17

Demonstrates: dept-hint override with justification, rating rubric application, TLDR form, Inbox bullet format.

Triage v0 Spec — /triage-inbox (lite)