nemawashi-update

name	nemawashi-update
description	Use when user wants to refresh one or more profiles end-to-end - runs nemawashi-collect followed by nemawashi-analyze for each named target (or --all), dispatching one profile-updater agent per target with throttled parallelism.

Profile Updater

End-to-end profile refresh — runs nemawashi-collect followed by nemawashi-analyze for each named target in a single invocation. The two-step pipeline (collect → analyze) is the common case; this skill is the convenience wrapper that dispatches one profile-updater agent per target, batched to respect MCP rate limits.

The individual skills remain available for the cases where they don't compose — re-analyzing without re-collecting, or seeding a profile for a brand-new person who needs the interactive collect dialog.

When to Use

• User says "update X's profile" / "refresh X" / "X を update して"
• User says "update everyone" / "全員 update" / explicit --all
• User has multiple targets to refresh (this skill parallelizes them; invoking the two underlying skills sequentially serializes everything)

Prerequisites

• The underlying skills' prerequisites: at least one MCP source available for nemawashi-collect; skills/nemawashi-analyze/FRAMEWORKS.md registry intact for nemawashi-analyze.
• For --all, at least one existing profile under PROFILE_DIR/.

Surface

nemawashi-update <name>                    # single target
nemawashi-update <a> <b> <c>               # multiple targets
nemawashi-update --all                     # every profile under PROFILE_DIR

Empty invocation (no positional, no --all) → print usage and stop.

Process

Step 1: Parse targets

Resolve the target list:

• Positional arguments → each is one target name (matches a directory under PROFILE_DIR/, or a new name to be created by the nested collect step).
• --all → list ${PROFILE_DIR}/*/ and use each basename as a target. If the directory is empty (no profiles yet), print:

  "No profiles under PROFILE_DIR; run /supernemawashi:nemawashi-collect <name> to seed one first." and stop.

--all is eager — no "Update N people, proceed?" confirmation. The user has full Ctrl-C to abort.

Step 2: Dispatch profile-updater agents

For each target, build the dispatch prompt:

profile_name:    <slug>
profile_dir:     ${PROFILE_DIR}/<slug>/
today:           <YYYY-MM-DD>

Issue dispatches in batches of 5 concurrent agents. Wait for the batch to return before dispatching the next batch. This top-level throttle is what keeps the nested profile-collector dispatches under the documented MCP rate-limit batch size (the local-only framework-analyzer fan-out inside each profile-updater is not the bottleneck).

For a single target (N=1), dispatch one agent and skip the batching logic. For N ≤ 5, all in one batch.

Step 3: Aggregate per-target reports

After every batch returns, collect the per-target report blocks. Each is one of:

<profile_name>: success
  collect: <facts_added N from M adapters>
  analyze: <produced N framework files; core pattern: "<one-line>"; contradictions: <count>>

<profile_name>: partial
  collect: <ran, no facts gathered | facts_added N from M adapters>
  analyze: <skipped | produced N framework files with K Data Gap>

<profile_name>: failed
  collect: <reason>
  analyze: skipped

Step 4: Print summary

Print a final aggregated table:

| Target | Status | Notes |
|---|---|---|
| <name> | success | <one-line summary from the report> |
| <name> | partial | <reason> |
| <name> | failed  | <reason> |

Followed by a one-line totals row: Total: N successful, M partial, K failed.

If any target failed or was partial, suggest follow-up actions:

• failed (collect) → "Check MCP authentication; try /supernemawashi:nemawashi-collect <name> directly to see the interactive error."
• partial (no facts) → "No new signals from any MCP source — the target may be a stale identifier."

Key Principles

• Compose, don't reinvent. This skill never duplicates collect or analyze logic; both flow through their canonical worker agents.
• Throttle at the top. The 5-batch cap is the only place rate limits are enforced; nested agents do not retry or back off.
• Failures are local. One target's failure must not block the others.
• Eager --all. Confirmation is the user's responsibility (Ctrl-C); the skill optimizes for the common "yes, just run it" path.

Coexistence with other skills

• nemawashi-collect — still the right tool when you want the interactive create-a-new-profile dialog or when MCP is the only thing that needs to run.
• nemawashi-analyze — still the right tool for re-analysis without re-collection (cheap; no MCP calls).
• nemawashi-discover — runs before this skill when you don't yet know who to update. Discovery cache (#1) will short-circuit the nested collect when implemented.
• nemawashi-check — runs before this skill when you want to refresh only stale profiles. Pipe its output (or hand-pick) into this skill's positional args.