/flow-next-capture

name: flow-next-capture
description: Synthesize the current conversation context into a flow-next spec at `.flow/specs/<spec-id>.md` via `flowctl spec create + spec set-plan` — agent-native, source-tagged, with mandatory read-back before write. Triggers on /flow-next:capture, "capture spec", "lock down what we discussed", "make a spec from this conversation", "convert conversation to spec". Optional `mode:autofix` token runs without questions and requires `--yes` to commit. Optional `--rewrite <spec-id>` overwrites an existing spec; `--from-compacted-ok` overrides the incomplete-evidence refusal after compaction; `--override-strategy` proceeds despite a contradiction with an active STRATEGY.md track (and prompts to record the override as a decision).
user-invocable: false
allowed-tools: AskUserQuestion, Read, Bash, Grep, Glob, Write, Edit, Task

/flow-next:capture — agent-native conversation → spec

A free-form discussion (or a `/flow-next:prospect` survivor) frequently produces enough material for a complete spec, but stops short of the formal `flowctl spec create` + `spec set-plan` heredoc documented in `CLAUDE.md`. Without an explicit synthesis step, that context decays — the next session loses the conversation, the spec never lands, and the user re-explains the same idea to `/flow-next:plan`.

This skill IS the synthesis. The host agent (Claude Code / Codex / Droid) extracts the recent user turns, drafts a CLAUDE.md-shaped spec with **per-line source tags** (`[user]` / `[paraphrase]` / `[inferred]` / `[strategy:<track>]`), **prints the full draft as ordinary markdown then issues a short approval ask** (print-then-ask — never embed multi-paragraph drafts in `AskUserQuestion` bodies), and only then writes the spec via existing flowctl plumbing. There is no Python synthesizer, no codex / copilot subprocess, no fast-model classifier. The host agent is already an LLM and does the work directly.

flowctl provides thin spec plumbing (`spec create`, `spec set-plan`, optional `spec set-branch`, `memory search` for duplicate detection) plus the chart handoff callback (`chart link-spec`) after a successful chart-briefing capture. Capture never writes chart files and never mutates a chart's `ready` flag; chart never writes `.flow/specs`.

Routing boundary (fn-135 / guide matrix)

Clear meaningful ideas and finished chart briefings route **here** - to capture (or direct spec authoring). Capture does **not** manufacture a chart for clear work. When intent and boundaries are already stateable, skip chart (`signal absent`). After a structured brief lands, narrow or skip interview only once read-back proves no material gaps - never pre-skip interview on hope. Unsure: `/flow-next:guide`.

Chart briefing ingestion (fn-135)

When the conversation (or `$ARGUMENTS`) references a chart briefing — a path under `.flow/charts/*-briefing*.md`, an explicit B-ID (`B1`, `B2`, …), or a chart id with a published briefing — capture treats that briefing as **attributable evidence**, not as pre-tagged acceptance criteria:

1. **Detect** the briefing input early (Phase 1 evidence). Read the index (and cluster file when multi-spec). Record chart id, B-ID, cluster key (if any), D-ID links, and approved asset references. 2. **Admission (fail closed):** ordinary capture **REFUSES draft or stale briefings**. A forced draft (`status: draft`) is never treated as final. A stale B-ID (after `chart reopen` or supersession of linked D-IDs) is refused by default. 3. **Explicit risk override only:** to admit a draft or stale briefing, the user must name the unresolved or invalidated D-IDs and the agent must **read back the exact risk** before write. The override never promotes a forced draft into a final briefing and never rewrites chart history. 4. **Provenance separation (load-bearing):**

• Chart/B-ID/cluster/D-ID evidence and approved assets go into `## Decision Context` / evidence sections as **links and references** — never with trailing `[user]` / `[paraphrase]` / `[inferred]` / `[strategy:<track>]` tags.
• The four source tags apply **only** to acceptance criteria capture **newly authors**. Never retag existing criteria. A criterion derived from an unattended resolved D-ID is **not** automatically `[user]`.
• Do **not** introduce verified/inferred fact or decision grammar (fn-148 closed STOPPED — no verdict; it licenses nothing here).

5. **Write order after approval:** `spec create` → `spec set-plan` → `flowctl chart link-spec <chart> --briefing <B> --spec <S> --decisions <D,...> [--cluster <k>]`. Call `link-spec` **only after** each successful spec creation. Decline records nothing and leaves the chart resumable. 6. **Retry / partial multi-spec:** on retry, first check `produced_specs[]` (and existing specs) for this B-ID+cluster identity; if a link already exists, link/use that spec instead of minting a duplicate. Partial multi-spec capture records only successful links and resumes the failed cluster without duplicating the first. Shared-context D-IDs stay attributable in each handoff but become acceptance requirements only where read-back confirms the target spec needs that guarantee.

**Read [workflow.md](workflow.md) for the full phase-by-phase execution. Read [phases.md](phases.md) for the must-ask cases lookup, source-tag taxonomy, confidence tiers, and forbidden-behaviors list.**

Preamble

**CRITICAL: flowctl is BUNDLED — NOT installed globally.** `which flowctl` will fail (expected). Define once; subsequent blocks (here and in `workflow.md` / `phases.md`) use `$FLOWCTL`:

FLOWCTL="${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/scripts/flowctl"
[ -x "$FLOWCTL" ] || FLOWCTL=".flow/bin/flowctl"

**Inline skill (no `context: fork`)** — `AskUserQuestion` must stay reachable across phases. Subagents can't call blocking question tools (Claude Code issues #12890, #34592). Phase 0 (duplicate detection) and Phase 4 (read-back loop) both require user choice in interactive mode. (sync-cod