flow-next-prospect

name	flow-next-prospect
description	Generate ranked candidate ideas grounded in the repo, upstream of /flow-next:plan. Triggers on /flow-next:prospect with an optional focus hint (concept, path, constraint, or volume).
user-invocable	false
allowed-tools	AskUserQuestion, Read, Bash, Grep, Glob, Write, Edit, Task

Prospect — upstream-of-plan idea generation

Read workflow.md for full phase-by-phase execution.

Generate many candidate ideas grounded in the repo, critique every one with explicit rejection reasons, and surface only the survivors bucketed by leverage. Output is a ranked artifact under .flow/prospects/<slug>-<date>.md that feeds directly into /flow-next:interview or /flow-next:plan via flowctl prospect promote.

Role: idea-prospecting coordinator (sequential single-chat — generate → critique → rank → write → handoff). Personas are prompt-level scaffolding inside this skill, not parallel subagent dispatch.

Preamble

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

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

Inline skill (no context: fork) — keeps AskUserQuestion available throughout. Subagents can't call blocking question tools (Claude Code issues #12890, #34592), and Phases 0 + 6 both require user choice. (sync-codex.sh rewrites this to a plain-text numbered prompt in the Codex mirror.)

Input

Arguments: $ARGUMENTS

Format: [focus hint] — freeform single string. Optional. May be:

• Concept — DX improvements, review-skill polish, test-suite health
• Path — plugins/flow-next/skills/ (ideate inside a subtree)
• Constraint — quick wins under 200 LOC, minor-bump only, no new deps
• Volume hint — top 3 (exactly 3 survivors), 50 ideas (generate ≥50), raise the bar (60-70% rejection target)

If empty, the skill picks its own coverage targets (15-25 candidates → 5-8 survivors).

Ralph-block (R8)

/flow-next:prospect is exploratory and human-in-the-loop. Autonomous loops have no business deciding what a repo should tackle next — that's a judgement call. Hard-error with exit 2 when running under Ralph.

if [[ -n "${REVIEW_RECEIPT_PATH:-}" || "${FLOW_RALPH:-}" == "1" ]]; then
  echo "Error: /flow-next:prospect requires a user at the terminal; not compatible with Ralph mode (REVIEW_RECEIPT_PATH or FLOW_RALPH detected)." >&2
  exit 2
fi

No env-var opt-in. Ralph never decides direction.

Workflow

Execute the phases in workflow.md in order:

0. Resume check — list active artifacts <30d; ask extend / fresh / open via blocking question. Corrupt artifacts surfaced but never offered for extension.
1. Ground — scan repo with graceful degradation: git log (30d), open specs, CHANGELOG top, memory matches, memory audit (if present), strategy snapshot (verbatim name / target_problem / approach / tracks / last_updated from flowctl strategy read --json when sections_filled >= 1; husk-vs-presence gate uses sections_filled, NOT [[ -f STRATEGY.md ]]). Emit a structured 30-50 line snapshot — titles + tags only, never raw bodies.
2. Generate — divergent-convergent + persona seeding (≥2 of senior-maintainer / first-time-user / adversarial-reviewer, picked by focus hint per personas.md). One divergent prompt; no self-judging.
3. Critique — separate prompt pass that does NOT see the focus hint or persona texts; rejection floor ≥40% (≥60% under raise the bar); fixed taxonomy (duplicates-open-epic | out-of-scope | out-of-scope-vs-strategy | insufficient-signal | too-large | backward-incompat | other); out-of-scope-vs-strategy is advisory only (user can override at promote time via existing --force flag); floor violation surfaces blocking question with frozen options regenerate | loosen-floor | ship-anyway.
4. Rank — bucketed: high leverage 1-3, worth-considering 4-7, if-you-have-the-time 8+. Forced-format leverage sentence per survivor (Small-diff lever because X; impact lands on Y.); no numeric scores.
5. Write artifact — atomic write-then-rename to .flow/prospects/<slug>-<date>.md via flowctl.write_prospect_artifact. Same-day collisions suffix with -2, -3. Optional floor_violation / generation_under_volume flags round-trip when upstream phases set them.
6. Handoff — blocking prompt for promote / interview / skip via the platform's question tool; frozen numbered-options fallback when no blocking tool is available.

Phases 0-6 are implemented. Promote command + list/read/archive land in tasks 4-5.

Pre-check: Local setup version

Compare .flow/meta.json setup_version to the plugin version; on mismatch, escalate once per plugin version. Fail-open throughout: a missing jq, .flow/meta.json, or plugin manifest silently continues.

SETUP_MODE=$(jq -r '.setup_mode // empty' .flow/meta.json 2>/dev/null)
SETUP_VER=$(jq -r '.setup_version // empty' .flow/meta.json 2>/dev/null)
PLUGIN_JSON="${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/.claude-plugin/plugin.json"
PLUGIN_VER=$(jq -r '.version' "$PLUGIN_JSON" 2>/dev/null || echo "unknown")
VERSION_ACK=$(jq -r '.version_ack // empty' .flow/meta.json 2>/dev/null)
if [[ "$SETUP_MODE" == "plugin" ]]; then
  # fn-121 plugin mode: no local copies exist to go stale - the version compare is
  # moot. Check only the CLAUDE.md snippet contract (sentinel vs the plugin's
  # expected v1; keep the literal in sync with SNIPPET_SCHEMA_VERSION in flowctl.py).
  SNIP_ACK=$(jq -r '.snippet_ack // empty' .flow/meta.json 2>/dev/null)
  SNIP_VER=$(grep -m1 -o 'flow-next:snippet:v[0-9]*' CLAUDE.md 2>/dev/null | grep -o '[0-9]*$')
  if [[ "${SNIP_VER:-missing}" != "1" ]]; then
    if [[ "${FLOW_RALPH:-}" == "1" || -n "${REVIEW_RECEIPT_PATH:-}" || "${FLOW_AUTONOMOUS:-}" == "1" || "${ARGUMENTS:-}" == *mode:autonomous* \
          || "$SNIP_ACK" == "1" ]]; then
      echo "CLAUDE.md flow-next snippet contract v${SNIP_VER:-missing} != plugin v1. Refresh via /flow-next:setup or the interactive ask." >&2
    else
      echo "FLOW_SNIPPET_ASK ${SNIP_VER:-missing} 1"
    fi
  fi
elif [[ -n "$SETUP_VER" && "$PLUGIN_VER" != "unknown" && "$SETUP_VER" != "$PLUGIN_VER" ]]; then
  if [[ "${FLOW_RALPH:-}" == "1" || -n "${REVIEW_RECEIPT_PATH:-}" \
        || "${FLOW_AUTONOMOUS:-}" == "1" || "${ARGUMENTS:-}" == *mode:autonomous* \
        || "$VERSION_ACK" == "$PLUGIN_VER" ]]; then
    echo "Local setup v${SETUP_VER} differs from plugin v${PLUGIN_VER}. Run /flow-next:setup to refresh local scripts." >&2
  else
    echo "FLOW_SETUP_ASK ${SETUP_VER} ${PLUGIN_VER}"
  fi
fi

If the block printed a FLOW_SNIPPET_ASK line (plugin mode only; suppressed to the stderr note under the autonomy markers above), before proceeding ask the user with AskUserQuestion (the CLAUDE.md flow-next snippet block is on an older contract than this plugin version; refresh the marker block?), offering exactly the options Refresh now, Remind me next version, Skip this run, then continue the skill whichever is chosen:

• Refresh now: run "${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/scripts/flowctl" setup-block apply --file CLAUDE.md --template "${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/skills/flow-next-setup/templates/claude-md-snippet-plugin.md" --json; if it returns action: ask, re-run as setup-block resolve with the same --file/--template plus --choice overwrite --json - this question WAS the consent. Marker-bounded: content outside the block is never touched.
• Remind me next version: record the acknowledgement so this contract version is not re-asked (fail-open: on any error, continue anyway):

  rm -f .flow/meta.json.tmp && jq '.snippet_ack = "1"' .flow/meta.json > .flow/meta.json.tmp && mv .flow/meta.json.tmp .flow/meta.json
  

• Skip this run: continue without writing anything; the next invocation asks again.

If the block printed a FLOW_SETUP_ASK line, before proceeding ask the user with AskUserQuestion (local setup differs from the plugin; refresh now?), offering exactly the options Refresh now, Remind me next version, Skip this run, then continue the skill whichever is chosen:

• Refresh now: pause and have the user run /flow-next:setup in this session (do not run setup yourself), then continue once it finishes.
• Remind me next version: record the acknowledgement so this version is not re-asked (only a later plugin version re-arms it), then continue. Run this self-contained write (fail-open: on any error, continue anyway):

  PJ="${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/.claude-plugin/plugin.json"
  PV=$(jq -r '.version' "$PJ" 2>/dev/null)
  [[ -n "$PV" && "$PV" != "null" ]] && rm -f .flow/meta.json.tmp && jq --arg v "$PV" '.version_ack = $v' .flow/meta.json > .flow/meta.json.tmp && mv .flow/meta.json.tmp .flow/meta.json
  

• Skip this run: continue without writing anything; the next invocation asks again.

Any other output (the one-line differs notice, or nothing) is non-blocking: continue.

Forbidden

• Running under Ralph — hard-block via the guard above.
• Setting context: fork — blocking question tools must stay reachable.
• Network calls — grounding is local-filesystem only (git, flowctl, memory, CHANGELOG).
• Writing to .flow/specs/ directly — only flowctl prospect promote may do that.
• Auto-archiving artifacts — only the explicit prospect archive subcommand moves files.
• Dumping raw file bodies into the grounding snapshot — titles + tags only; structured 30-50 lines max.