Role

Read-only. Render state.json plus a live scan of the work-queue directories into a compact, scannable status block, then state the single next action. This skill NEVER writes to state.json, the queue dirs, or anything else. It is safe to run at any moment, including mid-wave while workers are claiming units.

Invocation

/claude-migrate:progress [<run-name>] [--verbose]

• <run-name> optional. If omitted, auto-detect: list .planning/claude-migrate/ and if exactly one run dir exists, use it; otherwise print the available runs and exit.
• --verbose adds the optional detail block (Step 5).

Protocol

Step 1: Locate the run

• Parse <run-name> from $ARGUMENTS (the first token that is not a flag).
• If absent: ls .planning/claude-migrate/ - if exactly one dir, use it; else print the available runs and exit.
• RUN_PATH=".planning/claude-migrate/<run-name>".
• If $RUN_PATH/state.json is missing, STOP: "No such run. Initialize with /claude-migrate:init <run-name> first, or list runs with ls .planning/claude-migrate/."

Step 2: Read state

jq . "$RUN_PATH/state.json"

Pull the fields you will render: run, created_at, updated_at, current_step, status, blocked_reason, input.mode, output.mode, output.user_chose_auto, output.browser.transport, output.browser.authed, the four gates.*.verdict (and gates.filter.user_confirmed), every counters.*, cost_estimate, and last_checkpoint.

Step 3: Cross-check counters against the filesystem (read-only)

The directory a unit lives in IS its state, so a quick scan corroborates the counters without trusting them blindly. Count, do not move:

for q in pending in-progress done failed dropped; do
  printf 'units/%-12s %s\n' "$q" "$(ls "$RUN_PATH/units/$q" 2>/dev/null | wc -l | tr -d ' ')"
done
for q in pending in-progress done failed; do
  printf 'seed/%-12s %s\n' "$q" "$(ls "$RUN_PATH/seed/$q" 2>/dev/null | wc -l | tr -d ' ')"
done
ls "$RUN_PATH/briefs"/*.brief.md 2>/dev/null | wc -l   # distilled briefs on disk

If a filesystem count disagrees with the matching counter in state.json, note it as a one-line "counter drift" hint and point the user at /claude-migrate:health (which is the skill that diagnoses and proposes a repair). Do NOT attempt to reconcile here.

Step 4: Format output

claude-migrate run: <run>
  created: <created_at>    updated: <updated_at>
  input:   <input.mode>    output: <output.mode> (user_chose_auto=<bool>)
  browser: transport=<transport|none>  authed=<bool>
  status:  <status>        step: <current_step>   <blocked_reason if status==blocked>

Gates:
  pre-split: <verdict>      filter: <verdict> (user_confirmed=<bool>)
  verify:    <verdict>      pre-apply: <verdict>

Counters:
  chats     total=<chats_total>  kept=<kept>  dropped=<dropped>
  preflight pending=<P> in-progress=<IP> done=<D> failed=<F>
  distill   pending=<P> in-progress=<IP> done=<D> failed=<F>   (briefs ok=<briefs_verified_ok> fail=<briefs_verified_fail>)
  routing   seeded_units=<S>  doc_only_units=<DO>              (kept == seeded_units + doc_only_units)
  seed      pending=<P> in-progress=<IP> done=<D> failed=<F>   seeded=<seeded> renamed=<renamed> ok_protocol_miss=<M>
  projects  total=<T> pending=<P> created=<C> finalized=<Fz>

Cost estimate: in=<in_tokens> out_est=<out_tokens_est>  $<usd_low>-$<usd_high>  (blend=<model_blend|n/a>)
Filesystem cross-check: <"matches counters" | "DRIFT: <one-line> - run /claude-migrate:health">
Last checkpoint: <last_checkpoint|none>

Next action:
  <step-specific guidance - see the table below>

Show only the queues that apply to the run: the seed/projects lines and the browser line are meaningful only when output.mode == "auto"; for copy-page mode print "(copy-page mode - no browser seeding)" in their place.

Next-action guidance by current_step + status

If counters.kept == 0 at ready/done, replace the next-action line with the prominent Edge-case message: "0 chats kept - nothing to migrate. Review the DROP list in <RUN_PATH>/units/dropped/ and the per-unit reasons in <RUN_PATH>/value/." Never present an empty run as a silent success.

Step 5: Optional detail (only if --verbose)

• Last 20 lines of <RUN_PATH>/run.log (already pre-redacted at write time).
• Failed units: ls "$RUN_PATH/units/failed" (basenames only).
• Dropped units with their reason: read each value/<UNNN>.value.json .reason for units now in units/dropped/.
• Duplicate clusters surfaced at preflight: any value/*.json carrying looks_duplicate_of.
• Seed units not yet renamed: scan seed/*/UNNN.json for status in {seeded, awaited_ok}.

Hard rules

• Read-only. NEVER mutate state.json, the queue dirs, the run config, or any artifact. No state.sh set/inc/dec, no mv, no rm.
• Always print the "Next action" line. The user should never have to guess what to do next.
• Never invent a counter, gate, or step that is not in state.json / the §3.1 enum.
• Never display secrets or PII. Render only hashes and counts; never echo dest_chat_url, raw briefs, emails, or tokens. The run.log tail is already pre-redacted - do not un-redact it.
• Never reference any specific domain; bucket and group labels come from config.yaml (bucket_labels), not from this skill.