flow-next-sync — Claude Codex skill: sync specs after code drift
flow-next-sync runs plan-sync by hand to bring downstream task specs back in line once the code has moved ahead, touching tasks still in todo or blocked.
MIT ★ 730 updated 2026-10-29
/plugin marketplace add gmickel/flow-next
/plugin install flow-next@flow-nextWhat flow-next-sync does
This skill calls plan-sync on demand. Give it a task id like fn-N-slug.M or a spec id like fn-N-slug, and add --dry-run when you want the list of edits without writing them. Tracker handles such as wor-17 resolve too, because flowctl show maps them to the linked spec or task.
First it reads references/working-rules.md. flowctl comes bundled with the plugin rather than being installed globally, so the skill locates it at the plugin root and falls back to .flow/bin/flowctl. In spec mode the anchor is the newest done task, or an in_progress one when nothing is finished yet; the remaining todo and blocked tasks count as downstream. Project glossary terms, active decisions and STRATEGY.md are written to a temp directory under .flow/tmp/ so the agent can spot renamed terms and conflicts with the strategy.
A .flow/ directory has to exist before anything runs: flowctl init comes first, otherwise the skill exits with an error. The cross-spec phase only happens when planSync.crossSpec is true, and without that flag it is skipped entirely.
What it helps with
- pull downstream todo specs back in line after the code moved
- preview plan-sync edits with
--dry-runbefore anything is written - anchor drift detection on the newest finished task
- catch a renamed term against the project glossary
- check downstream tasks against
STRATEGY.md
Example prompts for the agent
- Sync the downstream task specs after my code changes:
fn-1-add-oauth.2. - Show me what plan-sync would change for
fn-1-add-oauthwithout writing — run it with--dry-run. - The implementation drifted from the spec, update the todo tasks under
wor-17.
How to install flow-next-sync
1Plugin marketplace
/plugin marketplace add gmickel/flow-next/plugin install flow-next@flow-nextThis installs the whole flow-next plugin with all its skills
2Skills CLI (Vercel)
npx skills add gmickel/flow-next --skill flow-next-sync -a claude-code3Manually: put the files in place
~/.claude/skills/flow-next-sync/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/flow-next-sync/
1Skills CLI (Vercel)
npx skills add gmickel/flow-next --skill flow-next-sync -a codex2Manually: put the files in place
.agents/skills/flow-next-sync/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add gmickel/flow-next --skill flow-next-sync -a cursor2Manually: put the files in place
.cursor/skills/flow-next-sync/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install gmickel/flow-next plugins/flow-next/skills/flow-next-sync2Manually: put the files in place
.github/skills/flow-next-sync/SKILL.md1Skills CLI (Vercel)
npx skills add gmickel/flow-next --skill flow-next-sync -a gemini-cli2Platform CLI
gemini skills install https://github.com/gmickel/flow-next --path plugins/flow-next/skills/flow-next-syncManually: put the files in place
.devin/skills/flow-next-sync/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about flow-next-sync
How to install flow-next-sync in Claude Code?
Use the plugin marketplace: /plugin marketplace add gmickel/flow-next, then /plugin install flow-next@flow-next. The skill arrives with the flow-next plugin.
What does flow-next-sync do?
It triggers plan-sync manually and rewrites the specs of downstream tasks that are still todo or blocked. Useful when code changes outpace the plan.
What does the skill need to run?
A .flow/ directory in the repo, created by flowctl init. The id you pass must resolve through flowctl show, or the run stops with an unknown-id message.
Source text
SKILL.md · MIT
---
name: flow-next-sync
description: Manually trigger plan-sync to update downstream task specs after implementation drift. Use when code changes outpace specs.
user-invocable: false
---
# Manual Plan-Sync
Manually trigger plan-sync to update downstream task specs.
Read [working-rules.md](../../references/working-rules.md) first unless you already have this run; it holds for every step of this skill.
## Preamble
**CRITICAL: flowctl is BUNDLED - NOT installed globally.** Define once; subsequent blocks use `$FLOWCTL`:
```bash
FLOWCTL="${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/scripts/flowctl"
[ -x "$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"
```
## Input
Arguments: $ARGUMENTS
Format: `<id> [--dry-run]`
- `<id>` - task ID `fn-N-slug.M` (or legacy `fn-N.M`, `fn-N-xxx.M`) or spec ID `fn-N-slug` (or legacy `fn-N`, `fn-N-xxx`), **or a resolvable tracker handle** (`wor-17` / `wor-17.M`) that `flowctl show` maps to the linked spec/task
- `--dry-run` - show changes without writing
## Workflow
### Step 1: Parse Arguments
```bash
REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)"
```
Parse $ARGUMENTS for:
- First positional arg = `ID`
- `--dry-run` flag = `DRY_RUN` (true/false)
**Validate ID first (handle-recognition rule):**
- **The id is resolved by `flowctl show`, not by a prefix check.** A session that rejects a resolvable tracker handle as an unknown id — because it gated on "must start with `fn-`" — has broken this. Route the arg through `$FLOWCTL show <ID> --json` (Step 3); flowctl's widened resolver maps a tracker key (`wor-17` / `wor-17.M`) to its linked spec/task, so a resolvable handle is the existing spec/task, never a new id. `/flow-next:sync wor-17` therefore resolves the linked spec.
- If no ID provided: "Usage: /flow-next:sync <id> [--dry-run]"
- If the arg does not resolve via `flowctl show` (Step 3): "Unknown ID. Use fn-N-slug (spec) / fn-N-slug.M (task), a tracker handle (wor-17), or legacy fn-N, fn-N-xxx."
Detect ID type (use the canonical id from `flowctl show`):
- Contains `.` (e.g., fn-1.2, fn-1-add-oauth.2, wor-17.2) -> task ID
- No `.` (e.g., fn-1, fn-1-add-oauth, wor-17) -> spec ID
### Step 2: Validate Environment
```bash
test -d .flow || { echo "No .flow/ found. Run flowctl init first."; exit 1; }
```
If `.flow/` missing, output error and stop.
### Step 3: Validate ID Exists
```bash
$FLOWCTL show <ID> --json
```
If command fails:
- For task ID: "Task <id> not found. Run `flowctl list` to see available."
- For spec ID: "Spec <id> not found. Run `flowctl specs` to see available."
Stop on failure.
### Step 4: Find Downstream Tasks
**For task ID input:**
```bash
# Extract spec from task ID (remove .N suffix)
SPEC=$(echo "<task-id>" | sed 's/\.[0-9]*$//')
# Get all tasks in spec
$FLOWCTL tasks --spec "$SPEC" --json
```
Filter to `status: todo` or `status: blocked`. Exclude the source task itself.
**For spec ID input:**
```bash
$FLOWCTL tasks --spec "<spec-id>" --json
```
1. First, find a **source task** to anchor drift detection (agent requires `COMPLETED_TASK_IDS`):
- Prefer most recently updated task with `status: done`
- Else: most recently updated task with `status: in_progress`
- Else: error "No completed or in-progress tasks to sync from. Complete a task first."
2. Then filter remaining tasks to `status: todo` or `status: blocked` (these are downstream).
**If no downstream tasks:**
```
No downstream tasks to sync (all done or none exist).
```
Stop here (success, nothing to do).
#### Done when
- A source task is anchored — the input task in task mode, or the most recently updated `done` (else `in_progress`) task in spec mode — or the run stopped with the documented refusal because neither exists.
- `DOWNSTREAM_TASK_IDS` holds the spec's `todo` and `blocked` tasks with the source task excluded. **An empty downstream set stops the run here.** A session that spawns the agent with nothing downstream has broken this.
### Step 5: Gather glossary + decisions + strategy context
Three extra context types help the agent catch drift the spec text alone can't reveal: project-glossary terms (renames where the old spec used a term whose `_Avoid_` alias now appears in code), active decision constraints (current code may touch files mentioned in a decision's `Consequences` section), and strategic-intent drift (completed task contradicts an active `STRATEGY.md` track or approach).
Write the three inputs to files; dispatch only their paths. Empty or failed reads
use the same defaults as an absent source.
```bash
# fence:plan-sync-inputs — FLOWCTL is the resolved executable path
mkdir -p .flow/tmp
SYNC_CONTEXT=$(mktemp -d "$(pwd)/.flow/tmp/plan-sync.XXXXXX") || exit 2
GLOSSARY_JSON_FILE="$SYNC_CONTEXT/glossary.json"
DECISIONS_JSON_FILE="$SYNC_CONTEXT/decisions.json"
STRATEGY_CONTENT_FILE="$SYNC_CONTEXT/strategy.json"
if ! "$FLOWCTL" glossary list --json > "$GLOSSARY_JSON_FILE" 2>/dev/null || [ ! -s "$GLOSSARY_JSON_FILE" ]; then
printf '%s\n' '{"groups":[],"file_count":0,"total_terms":0}' > "$GLOSSARY_JSON_FILE"
fi
if ! "$FLOWCTL" memory list --track knowledge --category decisions --json > "$DECISIONS_JSON_FILE" 2>/dev/null || [ ! -s "$DECISIONS_JSON_FILE" ]; then
printf '%s\n' '{"entries":[],"legacy":[],"count":0,"status":"active"}' > "$DECISIONS_JSON_FILE"
fi
if ! "$FLOWCTL" strategy read --json > "$STRATEGY_CONTENT_FILE" 2>/dev/null || [ ! -s "$STRATEGY_CONTENT_FILE" ]; then
printf '%s\n' '{}' > "$STRATEGY_CONTENT_FILE"
fi
printf '%s\n' "$GLOSSARY_JSON_FILE" "$DECISIONS_JSON_FILE" "$STRATEGY_CONTENT_FILE"
```
Record the printed absolute paths for the dispatch; variables do not persist
across shell calls. The agent reads the files and applies its Phase 3b husk
short-circuit. Keep all three files through the agent's return.
#### Done when
- All three files exist, preserving the empty defaults on failure.
- The dispatch carries file paths, never embedded glossary, decision, or strategy JSON.
### Step 6: Spawn Plan-Sync Agent
Read the cross-spec flag first — the same single config-leaf read `/flow-next:work` performs, so a repo that opted into cross-spec propagation (`planSync.crossSpec=true`) gets the same behavior from a manual `/flow-next:sync` as from the work-loop auto-trigger. Without this, `CROSS_SPEC` is unset and plan-sync skips the cross-spec phase entirely — the tool you reach for after big drift silently checks only same-spec tasks:
```bash
CROSS_SPEC=$($FLOWCTL config get planSync.crossSpec --json | jq -r '.value')
```
Build context and spawn via Task tool:
```
Sync task specs from <source> to downstream tasks.
COMPLETED_TASK_IDS: <source task id - the input task, or selected source for spec mode>
FLOWCTL: <resolved executable path from the preamble>
SPEC_ID: <spec id>
DOWNSTREAM_TASK_IDS: <comma-separated list from step 4>
DRY_RUN: <true|false>
CROSS_SPEC: <the $CROSS_SPEC value read below — literal "true" or "false", NOT "true|false">
GLOSSARY_JSON_FILE: <absolute glossary.json path from step 5>
DECISIONS_JSON_FILE: <absolute decisions.json path from step 5>
STRATEGY_CONTENT_FILE: <absolute strategy.json path from step 5>
<if DRY_RUN is true>
DRY RUN MODE: Report what would change but do NOT use Edit tool. Only analyze and report drift.
</if>
```
Use Task tool with `subagent_type: flow-next:plan-sync`.
**Note:** `COMPLETED_TASK_IDS` is always provided - for task-mode it's the input task, for spec-mode it's the source task selected in Step 4.
#### Done when
- `CROSS_SPEC` was read from `planSync.crossSpec` and passed to the agent as the literal string `true` or `false`.
- **The downstream task-spec edits are made by the spawned `flow-next:plan-sync` agent.** A session that edits downstream task specs directly has broken this.
- Under `--dry-run` the agent is told to report drift without using Edit, and Step 7 closes with "No files modified."
### Step 7: Report Results
After agent returns, format output:
**Normal mode:**
```
Plan-sync: <source> -> downstream tasks
Scanned: N tasks (<list>)
<agent summary>
```
**Dry-run mode:**
```
Plan-sync: <source> -> downstream tasks (DRY RUN)
<agent summary>
No files modified.
```
## Error Messages
| Case | Message |
|------|---------|
| No ID provided | "Usage: /flow-next:sync <id> [--dry-run]" |
| No `.flow/` | "No .flow/ found. Run `flowctl init` first." |
| Unknown ID (does not resolve) | "Unknown ID. Use fn-N-slug (spec) / fn-N-slug.M (task), a tracker handle (wor-17), or legacy fn-N, fn-N-xxx." |
| Task not found | "Task <id> not found. Run `flowctl list` to see available." |
| Spec not found | "Spec <id> not found. Run `flowctl list` to see available." |
| No source (spec mode) | "No completed or in-progress tasks to sync from. Complete a task first." |
| No downstream | "No downstream tasks to sync (all done or none exist)." |
## Rules
- **Ignores config** - `planSync.enabled` setting is for auto-trigger only; manual always runs
- **Any source status** - source task can be todo, in_progress, done, or blocked
- **Includes blocked** - downstream set includes both `todo` and `blocked` tasks
- **Reuses agent** - spawns existing plan-sync agent, no duplication
© Gordon Mickel · MIT · gmickel/flow-next