gmickel / flow-next-strategy

/flow-next:strategy — repo-root STRATEGY.md anchor

flow-next-strategy produces and maintains STRATEGY.md — a short, durable anchor at the repo root (peer of README.md / GLOSSARY.md) that captures what the product is, who it serves, how it succeeds, and where the team is investing. Downstream skills (/flow-next:prospect, /flow-next:plan, /flow-next:refine, /flow-next:capture, /flow-next:sync) read it as grounding when sections_filled >= 1.

The document is short and structured on purpose. Good answers to a handful of sharp questions produce a better strategy than any amount of prose. This skill asks those questions, pushes back on weak answers, and writes the doc.

Date the strategy document from date -u +%Y-%m-%d (run it; never assume the year).

Read [working-rules.md](../../references/working-rules.md) first unless you already have this run; it holds for every step of this skill.

Preamble

flowctl is bundled — NOT installed globally. which flowctl will fail (expected). Define once; subsequent blocks use $FLOWCTL:

FLOWCTL="${CODEX_HOME:-$HOME/.codex}/scripts/flowctl"
[ -x "$FLOWCTL" ] || FLOWCTL="<plugin-root>/scripts/flowctl"   # <plugin-root> = the directory two levels above this skill's SKILL.md file (the harness gave you that file's absolute path when the skill loaded); substitute it literally
[ -x "$FLOWCTL" ] || FLOWCTL=".flow/bin/flowctl"

Interaction Method

Ask the user via plain text. Render the options below as a numbered list 1. … N., followed by a final option N+1. Other — type your own answer. Print the question, then the numbered list, then stop and wait for the user's next message before continuing. Parse the reply as: a bare number 1–N+1 → that option; the literal text of an option label → that option; free text after Other → custom answer.

Default to plain-text numbered prompt. Never silently skip the question.

Free-form responses for the substantive sections (Target problem / Our approach / Who it's for / Key metrics / Tracks). Single-select with lead-with-recommendation only for routing decisions (which section to revisit, include this optional section, foreign-file resolution).

Focus Hint

<focushint> #$ARGUMENTS </focushint>

Interpret any argument as an optional focus: a section name to revisit (metrics, approach, tracks, problem, persona, milestones, not-working-on) or a scope hint. With no argument, proceed open-ended and let the file state decide the path.

Core Principles

1. Anchor, not plan. Strategy is what the product is and why. Features belong in /flow-next:prospect; tasks belong in specs and /flow-next:plan. Do not let either creep into the doc.
2. Rigor in the questions, not the headings. The section headers are plain English. The interview questions enforce strategy discipline (references/interview.md).
3. Short is a feature. The template is constrained. Adding sections costs more than it looks like. Push back on expansion.
4. Durable across runs. This skill is rerunnable. On a second run it updates in place, preserves what is working, and only challenges sections that look stale or weak.
5. Survives .flow/ wipe. STRATEGY.md lives at repo root, never under .flow/. The project's strategy belongs to the project, not flow-next.

Execution Flow

Phase 0: Route by file state

0.1 — Read file state

STATUS_JSON=$("$FLOWCTL" strategy status --json 2>/dev/null) || STATUS_JSON=
if ! printf '%s' "$STATUS_JSON" | jq -e '
  type == "object"
  and (.exists | type == "boolean")
  and (.husk | type == "boolean")
  and (.sections_filled as $n
    | ($n | type) == "number"
    and ($n | floor) == $n
    and $n >= 0
    and $n <= 7)
  and (.total_sections as $n
    | ($n | type) == "number"
    and ($n | floor) == $n
    and $n >= 5
    and $n <= 7)
  and (.sections_filled <= .total_sections)
  and (.last_updated == null or (.last_updated | type == "string"))
  and (.file_path == null or (.file_path | type == "string"))
  and (.generator == null or (.generator | type == "string"))
  and (.generator_match | type == "boolean")
  and (.husk == ((.exists == true) and (.sections_filled == 0)))
  and (.generator_match == (.generator == "flow-next-strategy"))
' >/dev/null 2>&1; then
  echo "[STRATEGY: unable to classify STRATEGY.md safely — leaving it unchanged]" >&2
  exit 0
fi
EXISTS=$(printf '%s' "$STATUS_JSON" | jq -r '.exists')
HUSK=$(printf '%s' "$STATUS_JSON" | jq -r '.husk')
SECTIONS_FILLED=$(printf '%s' "$STATUS_JSON" | jq -r '.sections_filled')
GENERATOR_MATCH=$(printf '%s' "$STATUS_JSON" | jq -r '.generator_match')
FILE_PATH=$(printf '%s' "$STATUS_JSON" | jq -r '.file_path // empty')

JSON fields (frozen by Task 1):

• exists (bool) — file present
• husk (bool) — exists: true AND sections_filled == 0
• sections_filled (int) — populated required + included optional-section count (0-7)
• total_sections (int) — 5 required + populated optional sections (5-7)
• last_updated (str|null) — ISO date from frontmatter
• file_path (str|null) — absolute path of resolved STRATEGY.md
• generator (str|null) — frontmatter generator value
• generator_match (bool) — generator == "flow-next-strategy"

0.2 — Subdirectory walk-up surfacing

If file_path is set and differs from ${PWD}/STRATEGY.md, surface one line in chat before any question fires:

Using repo-root STRATEGY.md at <file_path>.

This is the only line printed before routing — keep the noise floor low.

0.3 — Foreign-file resolution

If exists: true AND generator_match: false, do not write: read
[references/foreign-file.md](references/foreign-file.md) and ask its question before anything else.

0.4 — Routing

After walk-up surfacing and foreign-file resolution:

Announce the selected path, then load exactly one direct workflow reference:

• First-run: say Strategy doc not found — let's write it., then read and follow references/first-run.md.
• Update: say Found existing strategy — let's review and update., then read and follow references/update.md.

Do not read the unselected workflow. A foreign file stays entirely in Phase 0.3
unless the user confirms rewrite; confirmed rewrite selects the first-run
workflow. Any state not matched by the table is unsafe to classify: leave the
file unchanged and exit 0 with the same safe-classification stderr line from
Phase 0.1.

Done when: STATUS_JSON has validated, the selected path was announced, and exactly one of references/first-run.md / references/update.md has been read — or the run exited 0 leaving STRATEGY.md untouched.

Phase 3: Downstream handoff

After writing (first-run or update), surface the file's role to the user in one paragraph:

• If .flow/specs/ is empty (and any legacy .flow/epics/ is also empty) AND .flow/prospects/ is empty: Strategy doc written. Next, /flow-next:prospect [optional focus] generates ranked candidate ideas grounded in the strategy you just captured.
• If .flow/ is populated: Strategy doc written. Downstream skills (/flow-next:prospect, /flow-next:plan, /flow-next:refine, /flow-next:capture, /flow-next:sync) will read STRATEGY.md as grounding on next invocation.

One paragraph max. No follow-up questions.

Done when: STRATEGY.md is on disk at the repo root with the sections the interview filled, and exactly one handoff paragraph has been surfaced — nothing else printed at exit.

What this skill does not do

• Does not update the issue tracker or reconcile in-flight work. Strategy is the doc; execution lives in specs, tasks, and /flow-next:plan.
• Does not write product requirements or implementation plans — those are /flow-next:capture and /flow-next:plan.
• Does not compute metric values. It records which metrics matter and where they live, not what they read today.
• Does not create per-subdirectory STRATEGY.md files. Strategy is repo-wide by Rumelt's definition; cascading strategies re-introduce the "is for everyone, is for no one" problem.
• Does not migrate hand-written or CE-format STRATEGY.md files. v1 ships sentinel-based foreign-file refusal; multi-format migration is a v2 problem.
• Does not delete the file when all sections are removed. Last-section deletion leaves a husk (# <name> Strategy H1 + frontmatter) on disk — file never deleted.

Forbidden

• Setting context: fork — plain-text numbered prompt must stay reachable across phases.
• Inline cross-platform tool tables in prose (multi-platform listings naming the tool primitive on each harness).
• Lead-with-recommendation on substance questions — problem / approach / persona / metrics / tracks get free-form, no recommendation, no menu. Recommendation primes the user out of their own language. Routing questions only.
• Leaking anti-pattern names to the user. vanity / fluff / feature-list / goal-stated-as-problem are internal labels for formulating sharper follow-ups.
• Auto-overwriting a foreign-file STRATEGY.md — Phase 0.3 always asks. v1's stance is refusal; user can rename or delete to bootstrap.
• Writing more than 4 sentences per section (except Tracks, where each track has its own short block). The post-write checklist in references/strategy-template.md catches this.
• Adding sections beyond the locked 5 + 2 optional. CE's Marketing section is dropped on purpose; do not re-introduce it. Section order is locked.
• Inventing flowctl subcommands — The supported read surface is "$FLOWCTL" strategy {status,read} only. The skill writes the file directly via Write; no strategy add/list command exists.

Output rules

The deliverable is the written STRATEGY.md itself. Surface to chat:

• One-line path announcement at Phase 0 (walk-up subdir or file state).
• Per-section interview Q&A (the agent's questions; user's answers).
• Final draft read-back in Phase 1.5 / 2.4.
• One-paragraph downstream handoff at Phase 3.

No internal summary printed at exit beyond the Phase 3 handoff line. The file IS the report.