opencode-skills-collection 4.0.21 → 4.0.22

package/bundled-skills/shopify-review-triage/SKILL.md

@@ -0,0 +1,425 @@
+---
+name: shopify-review-triage
+description: "Turn public 1-3-star Shopify App Store review rows into a P0-P3 triage brief: incident risk, repeated friction, pricing confusion, feature requests, and an explicit needs-human-read bucket."
+category: product
+risk: none
+source: community
+source_repo: alfredtech2026/shopify-app-review-brief
+source_type: community
+date_added: "2026-08-03"
+author: alfredtech2026
+tags: [shopify, app-store-reviews, customer-feedback, triage, product-management, support]
+tools: [claude, cursor, codex, gemini, antigravity]
+license: "MIT"
+license_source: "https://github.com/alfredtech2026/shopify-app-review-brief/blob/main/LICENSE"
+---
+
+# Shopify Review Triage — public low-star reviews to a P0–P3 brief
+
+## Overview
+
+Takes rows of **public** Shopify App Store review text and produces one prioritized brief a
+product or support owner can act on: what kind of problem each review describes, how badly it
+can hurt, what to do first, and where the original wording came from.
+
+It is built for independent Shopify app teams and the agencies that run their support — the
+case where low-star reviews arrive scattered across several listings plus a few watched
+competitors, and the failure mode is treating them all as equally urgent.
+
+The rubric below is not invented here. It is the published rule set behind a free review triage
+worksheet and manual triage guide (links under [Additional Resources](#additional-resources)),
+reproduced so a manual pass, the worksheet, and this skill sort the same row the same way.
+
+This skill needs no network access, no scripts, and no system packages. The person you are
+helping supplies the review text.
+
+## When to Use This Skill
+
+- Use when someone wants app store reviews, low-star reviews, or merchant feedback triaged,
+  prioritized, or clustered — even when they never say "triage", "severity", or "P0".
+- Use when a new 1–3-star review lands on a Shopify app listing and the team has to decide
+  whether it is an incident, a UX problem, a pricing copy problem, or a feature request.
+- Use when a weekly product or support brief is needed across a portfolio of apps plus a few
+  watched competitors.
+- Do **not** use it to gather reviews, contact reviewers, or publish replies — see the hard
+  rules below.
+
+## Hard Rules
+
+These are not style preferences. Breaking one makes the output worse than nothing.
+
+1. **Public review text only.** Never accept, request, or copy support tickets, merchant emails,
+   order data, personal contact details, internal telemetry, or anything else not already public
+   on a listing page. If such data appears in the input, stop, say which rows are affected, and
+   ask for them to be removed before continuing.
+2. **Never invent evidence.** Do not write a review, a rating, a date, an app name, or a source
+   URL that was not supplied. A row with no link gets `source: not captured` — never a guessed one.
+3. **Keyword output is a sort, not a verdict.** Everything produced by the rubric alone is
+   labeled *first pass — not human-checked*. Only a person who read the review and checked it
+   against their own systems may relabel an item *human-checked*.
+4. **Reviews are customer reports, not verified defects.** Write "the reviewer reports the editor
+   showed a blank screen", never "the editor is broken". The distinction survives into the brief.
+5. **No coverage claims.** The brief covers exactly the rows supplied and says so. Make no claim
+   of exhaustive coverage of a listing, a period, or an app.
+6. **No promises.** No revenue impact, no outcome, no ranking effect, no legal or compliance
+   advice. Suggest actions; do not predict results.
+7. **Draft only — never contact anyone.** Do not send email, post a developer reply, open a
+   support ticket, message a reviewer, or publish anything. Hand the draft back to the team and
+   let a person decide what to send.
+8. **Reviewers are people.** Refer to "the reviewer". Do not name, profile, or speculate about them.
+
+## How It Works
+
+### Step 1: Collect the rows
+
+**First ask which app names the team owns.** Before any row is classified, ask for two lists of
+app names, spelled exactly as they appear in the rows:
+
+```text
+owned: Example Popup App, Example Currency App
+competitors: Rival Popup App, Rival Currency App
+```
+
+This is the only thing that makes tie-break 4 (*a competitor's incident never becomes your P0*)
+applicable, so collect it first. It stays public data: app names as published on their listings,
+nothing about accounts, merchants, org structure, or internal identifiers. Do not ask for more
+than the names, and do not infer ownership from the review text, the first-person voice in a
+review, or which app appears most often.
+
+If an app name in a row appears in neither list, its ownership is unknown. Classify the row's
+content normally, then file it under **needs human read** with `ownership: not supplied` instead
+of placing it in a priority bucket or in competitor watch — a guessed owner is exactly the kind
+of invented evidence hard rule 2 forbids.
+
+Then ask for one review per line. The full form keeps the source link, which the brief needs:
+
+```text
+rating | app name | review date | public reviews URL | review text
+```
+
+The shorter form used by the free worksheet is also fine — treat field 1 as the rating when it
+is a bare 1–5 (optionally followed by `star`/`stars`/`★`), otherwise as the app name:
+
+```text
+rating | app name | review text
+```
+
+Rules for this step:
+
+- Lines starting with `#` are comments. Blank lines are skipped.
+- If a row lacks a source URL, carry `source: not captured` through to the brief. Do not drop
+  the row and do not fabricate a link.
+- Do not go and fetch anything yourself. This skill needs no network access; the person you are
+  helping pastes the public rows they already opened.
+- The trigger this rubric is tuned for is a **new 1–3-star review**. Higher-rated rows still
+  classify correctly (a 5★ review often lands in feature requests or needs-human-read), so keep
+  them if they were supplied, but never present them as low-star signal.
+
+### Step 2: First pass — apply the rubric
+
+Lower-case the review text and normalize curly apostrophes (`’` → `'`) before matching, so a
+pasted "won’t load" still matches `won't load`.
+
+Five buckets. Each row gets exactly **one primary** bucket — the first dimension below, in this
+order, with any matching keyword. Further matches are recorded as **secondary**, never as a
+second brief item.
+
+#### P0 · Incident risk
+
+The purchase path, app activation, or merchant data may be at stake right now. Left alone it
+costs the merchant money and the team installs.
+
+**Suggested action.** Try to reproduce on a test store today. If confirmed, treat it as an incident: fix or mitigate first, then reply to the reviewer with what changed.
+
+**Signal keywords.** `won't load`, `wont load`, `won't open`, `wont open`, `can't close`, `cannot close`, `won't close`, `blank screen`, `broken`, `crash`, `stopped working`, `not working`, `doesn't work`, `does not work`, `checkout`, `losing sales`, `lost sales`, `error`
+
+#### P1 · Repeated friction
+
+The product works, but the same struggle keeps showing up across reviews or against an open
+support theme. Repetition is the signal, not volume of adjectives.
+
+**Suggested action.** Log it against the matching support theme. If the same complaint repeats across rows, schedule a UX fix ahead of new feature work.
+
+**Signal keywords.** `confusing`, `unclear`, `hard to`, `difficult`, `complicated`, `clunky`, `slow`, `couldn't figure`, `could not figure`, `annoying`, `had to contact support`, `setup took`, `too many steps`
+
+#### P2 · Pricing confusion
+
+What the merchant expected to pay and what happened diverged. Usually a copy problem in the
+listing, the plan limits, or the upgrade prompts — not a code problem.
+
+**Suggested action.** Compare what the reviewer expected with the listing's pricing section and in-app upgrade prompts; clarify the copy where they diverge.
+
+**Signal keywords.** `pricing`, `price`, `charged`, `charge`, `billing`, `billed`, `expensive`, `free plan`, `trial`, `refund`, `hidden fee`, `hidden cost`, `paywall`
+
+#### P3 · Feature request
+
+The merchant wants something the app does not do, or could not find. Valuable as a log entry,
+rarely urgent on its own.
+
+**Suggested action.** Add it to the feature-request log with a link to the review. If the capability already exists, reply to the reviewer with where to find it.
+
+**Signal keywords.** `wish`, `would be great`, `would love`, `please add`, `feature request`, `missing`, `if only`, `would like`, `no option to`, `needs an option`, `hope you add`, `add support for`
+
+#### Needs human read
+
+No keyword matched. Vague frustration, sarcasm, mixed praise, or a story that needs context.
+
+**Suggested action.** No keyword matched. Read the full review yourself and file it manually — the heuristic makes no guess here.
+
+**Priority.** The worksheet labels this bucket `P2` and sorts it last. Treat that label as
+provisional placement in the queue, not as a severity judgment — nothing has been judged yet.
+
+#### Tie-breaks and escalation
+
+1. **Most severe wins.** A row naming both a broken checkout and a billing surprise files under
+   P0 with pricing noted as secondary. Never split one review across two brief items.
+2. **Repetition escalates.** If the same friction or pricing theme appears in three or more
+   reviews within about 60 days, move it up one level and say how many rows drove the change.
+3. **Age discounts.** A review older than a year is background, not evidence of a current
+   problem, unless a recent row corroborates it. Cite it as context, never as the headline.
+4. **Competitor reviews never create a P0 for you.** Resolve the row's app name against the
+   ownership lists from step 1: `owned` keeps its rubric bucket, `competitors` moves to the
+   competitor watch section whatever its keywords matched, and a name in neither list goes to
+   needs human read with `ownership: not supplied`. A competitor's incident is roadmap,
+   positioning, or copy input — never your P0.
+5. **When unsure, choose needs human read.** The bucket exists so the rubric never launders
+   uncertainty into a priority label.
+
+### Step 3: Human pass — verify before you promote anything
+
+The first pass is where this skill stops being able to help on its own. Before any item is
+presented as more than a keyword match, a person on the team has to:
+
+- read the full original review at its source link;
+- for P0 candidates, attempt to reproduce on a development store and check the error tracker and
+  support inbox for matching signals from the same period;
+- record the outcome as *reproduced*, *not reproduced*, or *attempted — notes attached*.
+
+Ask for these outcomes rather than assuming them. Until you have them, every item stays labeled
+*first pass — not human-checked*, including in the summary line. An unverified P0 is a candidate,
+not an incident.
+
+Known limits to state plainly when they apply: keyword matching is English-only, misses sarcasm
+and context, can misfile a review that mentions "checkout" in passing, and sees only the rows
+supplied.
+
+### Step 4: Write the brief
+
+One document per portfolio, sections in rubric order, every item carrying an owner, a next
+action, and a source link. An item without an owner is a note, not a brief entry.
+
+<!-- brief-template -->
+```markdown
+# Low-star review brief — {portfolio or team name} — week of {YYYY-MM-DD}
+
+Scope: {apps monitored} · {competitors watched} · {N} rows supplied, {date range}.
+Covers only the rows supplied — no claim of exhaustive coverage.
+Reviews are customer reports, not verified defects. Items marked "first pass" are
+unverified keyword matches; "human-checked" means a person read the review and checked it.
+
+## P0 — Incident risk
+- **{App} — {signal in a few words}** ({rating}★, {review date}, source: {public reviews URL or not captured})
+  - Reviewer reports: {one sentence, in their words where possible}
+  - Status: first pass — not human-checked / human-checked
+  - Reproduced: {yes / no / attempted — notes}
+  - Next action: {action} — owner {name}, due {date}
+
+## P1 — Repeated friction
+- **{App} — {theme}** ({rating}★, {date}, source: {public reviews URL or not captured}; also seen: {where})
+  - Status: first pass — not human-checked / human-checked
+  - Next action: {UX or docs change} — owner {name}, due {date}
+
+## P2 — Pricing confusion
+- **{App} — {signal}** ({rating}★, {date}, source: {public reviews URL or not captured})
+  - Expected vs. actual: {one line}
+  - Status: first pass — not human-checked / human-checked
+  - Next action: {copy or prompt change} — owner {name}, due {date}
+
+## P3 — Feature requests
+- **{App} — {request}** ({rating}★, {date}, source: {public reviews URL or not captured}) — {log it / already exists → reply with where to find it}
+
+## Needs human read
+- **{App}** ({rating}★, {date}, source: {public reviews URL or not captured}) — {no keyword matched; what a human should look for}{, or: ownership: not supplied — app name on neither list}
+
+## Competitor watch
+- **{Competitor} — {signal}**: {what it implies for our roadmap, copy, or positioning}
+
+## Decisions this week
+- {one decision or experiment, with the row(s) that motivated it}
+```
+
+Open the summary line with the counts, e.g. *"Triaged 8 rows supplied: 3 incident risk,
+2 repeated friction, 1 pricing confusion, 1 feature request, 1 needs human read — first pass,
+not human-checked."*
+
+### Step 5: Self-check before you hand it over
+
+Refuse to deliver until every line is true:
+
+- [ ] Every item names its bucket and priority from the rubric above, and nothing else.
+- [ ] Every item carries a source link or an explicit `source: not captured`.
+- [ ] Every P0–P3 item is an app on the `owned` list; every competitor row sits in competitor
+      watch; every unlisted app name says `ownership: not supplied` under needs human read.
+- [ ] No review text, rating, date, app name, or URL appears that was not supplied.
+- [ ] Every unverified item says *first pass — not human-checked*; nothing claims a human check
+      that did not happen.
+- [ ] Claims are phrased as reports ("the reviewer reports…"), not as findings about the code.
+- [ ] The scope line says how many rows were supplied and makes no coverage claim.
+- [ ] No promise about revenue, ratings, outcomes, or compliance appears anywhere.
+- [ ] No private data survived into the output.
+- [ ] Nothing was sent, posted, or published — the brief is a draft for the team.
+
+## Examples
+
+### Example 1: Worked example — eight rows in, first pass out
+
+These eight fictional rows are the worksheet's own example set, so the two tools can be compared
+directly. Two of them are deliberately 4★ and 5★, to exercise the feature-request and
+needs-human-read buckets.
+
+Ownership context, collected before any of it is classified:
+
+```text
+owned: Example Popup App, Example Currency App, Example Reviews App
+competitors: (none supplied)
+```
+
+```text
+1 | Example Popup App | The editor shows a blank screen and the popup won't load. We are losing sales every day.
+2 | Example Popup App | The overlay can't close on mobile and it blocks the checkout button.
+1 | Example Currency App | Conversion is broken at checkout and we were still billed for the month.
+3 | Example Currency App | Setup took hours and the settings screen is confusing. Support was slow to reply.
+3 | Example Reviews App | The widget looks fine but the template editor is confusing and hard to use on a tablet.
+2 | Example Currency App | We kept getting charged after uninstalling, and the pricing page never mentioned this.
+4 | Example Reviews App | Great app, but I wish it could export reviews to CSV. Please add filtering by country.
+5 | Example Reviews App | Does what it promises and support replied the same day.
+```
+
+First pass over those rows:
+
+```text
+row 1 → P0 incident risk
+row 2 → P0 incident risk
+row 3 → P0 incident risk (secondary: pricing confusion)
+row 4 → P1 repeated friction
+row 5 → P1 repeated friction
+row 6 → P2 pricing confusion
+row 7 → P3 feature request
+row 8 → needs human read
+```
+
+**Explanation:** Rows 4 and 5 both matched `confusing`, so they are flagged as a repeated theme —
+two rows, which is a cluster to watch, not yet the three that trigger escalation. Row 3 is a
+single P0 item with pricing recorded as secondary, never two items. Row 8 matched nothing and
+stays unjudged. All three app names are on the `owned` list, so every bucket above is the team's
+own queue and competitor watch is empty; had `Example Reviews App` been listed as a competitor
+instead, rows 5, 7, and 8 would move there and none of them could become a P0. None of these rows
+carried a source URL, so each item would read `source: not captured` until the team supplies the
+listing links.
+
+### Example 2: A row that carries its source link
+
+```text
+1 | Example Popup App | 2026-07-28 | https://apps.shopify.com/example-popup-app/reviews?ratings%5B%5D=1 | The editor shows a blank screen and the popup won't load. We are losing sales every day.
+```
+
+Rendered into the brief:
+
+```markdown
+## P0 — Incident risk
+- **Example Popup App — editor reported blank, popup reported not loading** (1★, 2026-07-28, [source](https://apps.shopify.com/example-popup-app/reviews?ratings%5B%5D=1))
+  - Reviewer reports: the editor shows a blank screen, the popup does not load, and they are losing sales daily.
+  - Status: first pass — not human-checked
+  - Reproduced: not yet attempted
+  - Next action: attempt reproduction on a development store today — owner {name}, due {date}
+```
+
+**Explanation:** It files as a P0 only because `Example Popup App` is on the `owned` list; the
+same row from a competitor listing would render under competitor watch instead. The wording stays
+a report ("the reviewer reports"), the status stays *first pass — not human-checked* until a
+person verifies it, and the source link is the listing's public reviews page with the rating
+filter kept — the App Store has no per-review permalink.
+
+## Best Practices
+
+- ✅ **Do:** keep one review in exactly one bucket, and record extra matches as secondary notes.
+- ✅ **Do:** carry `source: not captured` forward when a row has no link, so the gap is visible.
+- ✅ **Do:** label every unverified item *first pass — not human-checked*, including in the summary.
+- ✅ **Do:** phrase every finding as a customer report, not as a confirmed defect.
+- ✅ **Do:** state how many rows were supplied and refuse any coverage claim beyond them.
+- ❌ **Don't:** fetch reviews, scrape listings, or ask for support tickets, emails, or order data.
+- ❌ **Don't:** invent a rating, date, app name, or URL that was not supplied.
+- ❌ **Don't:** send, post, or publish anything — including a developer reply to a reviewer.
+- ❌ **Don't:** promise a revenue, ratings, or compliance outcome from any suggested action.
+
+## Limitations
+
+- Keyword matching is **English-only**. Non-English reviews match nothing and land in
+  needs-human-read; that is the correct outcome, not a bug to work around by translating first.
+- The rubric misses sarcasm, irony, and context, and can misfile a review that mentions
+  "checkout" or "missing" in passing.
+- It sees only the rows the person supplies. It cannot know a listing's full review history, the
+  team's error tracker, or their support inbox.
+- It cannot verify anything. Every P0 it produces is a *candidate*, not a confirmed incident,
+  until a person reproduces it.
+- It does not replace environment-specific validation, testing, or expert review. Stop and ask
+  for clarification if required inputs, permissions, or safety boundaries are missing.
+
+## Security & Safety Notes
+
+- **No commands, no network, no credentials.** This skill runs on pasted text only. It must not
+  fetch listings, call APIs, or read files outside what the person supplies.
+- **Private data is a stop condition.** If support tickets, merchant emails, order records,
+  personal contact details, or internal telemetry appear in the input, stop, name the affected
+  rows, and ask for them to be removed before continuing.
+- **No outbound messaging.** The output is a draft handed back to the team. Sending email,
+  posting a public developer reply, opening a ticket, or contacting a reviewer is out of scope
+  under every circumstance (hard rule 7).
+- **Reviewers are people.** Do not name, profile, or speculate about a reviewer; refer to
+  "the reviewer".
+- **No promises.** No revenue, ratings, ranking, legal, or compliance claims belong in a brief.
+
+## Common Pitfalls
+
+- **Problem:** The Shopify App Store has no stable per-review permalink.
+  **Solution:** Cite the listing's public reviews page, keep the rating filter if one was used
+  (`…/reviews?ratings%5B%5D=1`), and pin the item with the review date plus the reviewer's first
+  few words so a human can find it again.
+- **Problem:** `checkout` is the noisiest keyword in the set — it fires on "we love the checkout
+  upsell".
+  **Solution:** A P0 whose only evidence is the word `checkout` is a needs-human-read row wearing
+  a P0 badge. Say so instead of promoting it.
+- **Problem:** `missing` and `error` cross buckets — "missing a dark mode" is P3, "settings page
+  errors out" is P0.
+  **Solution:** Primary-bucket order resolves the collision mechanically; the human pass fixes
+  the ones where it guessed wrong.
+- **Problem:** A competitor's incident looks worse than anything on the team's own listings.
+  **Solution:** It still goes to competitor watch. A competitor's P0 is never yours.
+- **Problem:** One review gets split across two sections, double-counting the same merchant and
+  inflating every count in the summary line.
+  **Solution:** One review, one item. Secondary matches are annotations.
+- **Problem:** The free in-browser worksheet parses three fields and folds everything after the
+  second `|` into the review text, so a five-field row displays its date and URL inside the quote.
+  **Solution:** Paste the short form into the worksheet and keep the long form here.
+
+## Related Skills
+
+- `@customer-research` — when the goal is broader voice-of-customer synthesis rather than
+  prioritizing a specific set of low-star review rows.
+- `@shopify-apps` — when the next step is actually building or fixing the Shopify app behavior a
+  triaged P0 points at.
+- `@before-you-build` — when a P3 feature request needs product-risk review before it becomes
+  roadmap work.
+
+## Additional Resources
+
+This skill packages the public rubric behind **Shopify App Review Brief**, an independent
+open-source project that is not affiliated with or endorsed by Shopify Inc. or any app developer.
+The same four dimensions, priorities, keyword lists, and suggested actions are published in three
+places:
+
+- [Manual guide, tie-break rules, and brief template](https://alfredtech2026.github.io/shopify-app-review-brief/guides/shopify-app-review-triage.html)
+- [Free in-browser worksheet that automates the first pass](https://alfredtech2026.github.io/shopify-app-review-brief/tools/review-triage-worksheet.html)
+- [Two worked sample briefs over real public reviews](https://alfredtech2026.github.io/shopify-app-review-brief/#samples)
+
+Upstream source repository: [alfredtech2026/shopify-app-review-brief](https://github.com/alfredtech2026/shopify-app-review-brief) (MIT).

package/bundled-skills/unified-ai-gateway/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: unified-ai-gateway
-description: Operate and evaluate Unified AI System through eight governed MCP tools while preserving fake-provider, authorization, and evidence boundaries.
+description: Operate and evaluate Unified AI System through nine governed MCP tools, including provider-free prompt enhancement, while preserving fake-provider, authorization, and evidence boundaries.
 category: ai-ml
 risk: critical
 source: https://github.com/happy520ai/unified-ai-system/tree/master/skills/unified-ai-gateway
@@ -27,20 +27,21 @@ installations require the manual setup below.
 ## Prerequisites And Setup
 
 1. Confirm that Codex CLI and Docker are installed and Docker is running.
-2. If the eight tools are already visible, skip setup and do not register a
+2. If the nine tools are already visible, skip setup and do not register a
    duplicate server.
-3. Explain the first stage: it downloads the reviewed immutable linux/amd64
-   image into Docker's cache, inspects its metadata and layer history, creates
-   but never starts a temporary container, exports its root filesystem, removes
-   that temporary container, and writes an inspection inventory to a temporary
-   directory. Obtain explicit user approval for those download and inspection
-   changes only.
+3. Explain the first stage: it downloads one reviewed platform from the
+   immutable `0.4.0` multi-platform index into Docker's cache, inspects its
+   metadata and layer history, creates but never starts a temporary container,
+   exports its root filesystem, removes that temporary container, and writes an
+   inspection inventory to a temporary directory. The reviewed platforms are
+   linux/amd64 and linux/arm64. Obtain explicit user approval for those download
+   and inspection changes only.
 4. After that first approval, pull the reviewed platform manifest and complete
    the inspection. Do not execute the image or register it yet:
 
 ```bash
-IMAGE='ghcr.io/happy520ai/unified-ai-system/mcp-server@sha256:cc17e923335f953631f59fb6a5ffcdce0e12e16c5abf362f1d28747452adadee'
-PLATFORM='linux/amd64'
+IMAGE='ghcr.io/happy520ai/unified-ai-system/mcp-server@sha256:c185d124d1f672b5cf210a7b7d4c7dbdc907b81a5f7b62fe312a0dc18839e045'
+PLATFORM='linux/amd64' # Use linux/arm64 only on a reviewed ARM64 engine.
 REVIEW_DIR="$(mktemp -d)"
 
 docker pull --platform "$PLATFORM" "$IMAGE"
@@ -55,13 +56,26 @@ tar -tf "$REVIEW_DIR/rootfs.tar" > "$REVIEW_DIR/rootfs-files.txt"
 mkdir -p "$REVIEW_DIR/rootfs"
 tar --same-permissions -xf "$REVIEW_DIR/rootfs.tar" -C "$REVIEW_DIR/rootfs"
 find "$REVIEW_DIR/rootfs/app" -type f -print > "$REVIEW_DIR/app-files.txt"
-find "$REVIEW_DIR/rootfs/app" \( -type l -o -type f -links +1 \) -exec ls -ld {} + > "$REVIEW_DIR/app-links.txt"
-find "$REVIEW_DIR/rootfs/app" -type f -name '*.node' -exec sha256sum {} + > "$REVIEW_DIR/native-binaries.sha256"
+: > "$REVIEW_DIR/app-links.txt"
+while IFS= read -r -d '' APP_LINK; do
+  ls -ld -- "$APP_LINK" >> "$REVIEW_DIR/app-links.txt"
+done < <(find "$REVIEW_DIR/rootfs/app" \( -type l -o -type f -links +1 \) -print0)
+: > "$REVIEW_DIR/native-binaries.sha256"
+while IFS= read -r -d '' NATIVE_BINARY; do
+  sha256sum -- "$NATIVE_BINARY" >> "$REVIEW_DIR/native-binaries.sha256"
+done < <(find "$REVIEW_DIR/rootfs/app" -type f -name '*.node' -print0)
 find "$REVIEW_DIR/rootfs" -type f \( -perm -0100 -o -perm -0010 -o -perm -0001 \) -print > "$REVIEW_DIR/executable-files.txt"
 find "$REVIEW_DIR/rootfs" -type f \( -perm -4000 -o -perm -2000 \) -print > "$REVIEW_DIR/suid-sgid-files.txt"
-find "$REVIEW_DIR/rootfs/app" -type f \( -name '.env' -o -name '.env.*' -o -name '*.pem' -o -name '*.key' -o -name '*.p12' -o -name '*.pfx' -o -name 'id_rsa*' \) -print > "$REVIEW_DIR/credential-like-files.txt"
-find "$REVIEW_DIR/rootfs/app" -type f -name package.json -exec grep -nHE '"(preinstall|install|postinstall|prepare|prepack|postpack)"' {} + > "$REVIEW_DIR/lifecycle-hooks.txt"
-grep -RInE 'child_process|spawn\(|fetch\(|AI_GATEWAY_MCP_URL|process\.env|writeFile|appendFile|unlink|rm\(' "$REVIEW_DIR/rootfs/app/packages/mcp-server/src" "$REVIEW_DIR/rootfs/app/packages/shared-sdk/src" > "$REVIEW_DIR/runtime-sensitive-code.txt"
+find "$REVIEW_DIR/rootfs/app" -type f \( -name '.env' -o -name '.env.*' -o -name '*.pem' -o -name '*.key' -o -name '*.p12' -o -name '*.pfx' -o -path '*/.ssh/id_*' \) -print > "$REVIEW_DIR/credential-like-files.txt"
+find "$REVIEW_DIR/rootfs/app" -type f -name 'package.json' \
+  -exec grep -nHE '"(preinstall|install|postinstall|prepare|prepack|postpack)"' -- {} + \
+  > "$REVIEW_DIR/lifecycle-hooks.txt"
+find \
+  "$REVIEW_DIR/rootfs/app/packages/mcp-server/src" \
+  "$REVIEW_DIR/rootfs/app/packages/shared-sdk/src" \
+  -type f \
+  -exec grep -nHE 'child_process|spawn\(|fetch\(|AI_GATEWAY_MCP_URL|process\.env|writeFile|appendFile|unlink|rm\(' -- {} + \
+  > "$REVIEW_DIR/runtime-sensitive-code.txt"
 ```
 
 If `sha256sum` is unavailable, use the platform's SHA-256 utility and preserve
@@ -70,19 +84,25 @@ deletion is another filesystem change and requires approval for the exact path.
 
 5. Read every generated inventory and report the inspection before proceeding.
    Compare it with the versioned
-   [image content review](https://github.com/happy520ai/unified-ai-system/blob/2967617bb266c79992a6ed40e70bb59e67f661ca/docs/security/mcp-image-review-0.3.2.md).
-   Require linux/amd64 manifest digest
-   `sha256:cc17e923335f953631f59fb6a5ffcdce0e12e16c5abf362f1d28747452adadee`,
-   config digest
-   `sha256:c28166511ca9d95ab6f4c0c8ac0cbc2acbbfdac20f971d074ff6744d39e0248f`,
-   source `https://github.com/happy520ai/unified-ai-system`, revision
-   `541430d68fac6b35c512ea7d2df20fe45334e0a5`, version `0.3.2`, license
+   [image content review](https://github.com/happy520ai/unified-ai-system/blob/4bbc5e81d1f372a5c80ba5597973f3284965adf6/docs/security/mcp-image-review-0.4.0.md).
+   Require OCI index digest
+   `sha256:c185d124d1f672b5cf210a7b7d4c7dbdc907b81a5f7b62fe312a0dc18839e045`.
+   For linux/amd64, require manifest digest
+   `sha256:bb3ba00366a924d511c776986f890d62196ecc380034daf9c42f54000dcc7f2d`
+   and config digest
+   `sha256:3224ec32c8a1407ba704febf897157866f6cabf86fb515d760b0466fe64c9df1`.
+   For linux/arm64, require manifest digest
+   `sha256:2a58da07d11de97a4b4051f4a82ac444e7fefb5235556d7997080c96db2da6ae`
+   and config digest
+   `sha256:1e480c2b6711283f9571079d96c73f5dfc423a30d86c22d05c0dfd052113a9b7`.
+   Require source `https://github.com/happy520ai/unified-ai-system`, revision
+   `9f606b0b4189ef9759bdc01857919c254209e4be`, version `0.4.0`, license
    `Apache-2.0`, entrypoint `docker-entrypoint.sh`, and command
    `node packages/mcp-server/src/index.js`.
 
    Report these reviewed risks explicitly: the image uses the default root
    user; includes Debian shell/package utilities and 11 base-image SUID/SGID
-   files; contains 549 internal pnpm links, two native Node binaries, and eight
+   files; contains 519 internal pnpm links, two native Node binaries, and eight
    lifecycle-hook declarations; and starts a child gateway with loopback HTTP.
    The optional `AI_GATEWAY_MCP_URL` can make an HTTP or HTTPS connection only
    when explicitly passed. The registered command below passes no host files,
@@ -98,12 +118,14 @@ deletion is another filesystem change and requires approval for the exact path.
    disabled, then inspect the stored configuration:
 
 ```bash
-codex mcp add unified-ai-system -- docker run --rm -i --pull never --platform linux/amd64 --network none --cap-drop ALL --security-opt no-new-privileges ghcr.io/happy520ai/unified-ai-system/mcp-server@sha256:cc17e923335f953631f59fb6a5ffcdce0e12e16c5abf362f1d28747452adadee
+IMAGE='ghcr.io/happy520ai/unified-ai-system/mcp-server@sha256:c185d124d1f672b5cf210a7b7d4c7dbdc907b81a5f7b62fe312a0dc18839e045'
+PLATFORM='linux/amd64' # Match the reviewed platform inspected above.
+codex mcp add unified-ai-system -- docker run --rm -i --pull never --platform "$PLATFORM" --network none --cap-drop ALL --security-opt no-new-privileges "$IMAGE"
 codex mcp get unified-ai-system --json
 ```
 
 8. Restart Codex or open a new task, then use `/mcp verbose` to confirm that all
-   eight tools are available. Remove the registration when it is no longer
+   nine tools are available. Remove the registration when it is no longer
    wanted:
 
 ```bash
@@ -141,6 +163,7 @@ deploying a production gateway.
 
 - `gateway_health`: managed gateway status and provider mode
 - `gateway_readiness`: chat-path readiness and blockers
+- `gateway_prompt_enhance`: local prompt structuring without a provider call
 - `gateway_chat`: deterministic credential-free chat proof
 - `knowledge_readiness`: knowledge subsystem readiness
 - `workflow_health`: workflow subsystem status
@@ -167,8 +190,8 @@ Agent:
 - Do not enable or call a real provider without explicit scoped authorization.
 - Treat MCP registration, image pulls, container creation, networking, and
   teardown as host-state changes that require informed user approval.
-- Never substitute a mutable tag, the multi-platform index, or an unreviewed
-  platform manifest for the reviewed linux/amd64 digest. Keep download and
+- Never substitute a mutable tag, a different OCI index, or an unreviewed
+  platform manifest for the reviewed `0.4.0` identities. Keep download and
   inspection approval separate from registration and activation approval.
 - Keep `--pull never` in the registered command. If the reviewed image is
   absent from the local cache, fail closed and return to the first approval
@@ -188,8 +211,8 @@ Agent:
 - The credential-free chat tool proves only the deterministic local fake path.
 - It does not configure real providers or handle provider credentials.
 - The published MCP image requires Docker.
-- The reviewed `0.3.2` path is limited to linux/amd64. Do not activate an arm64
-  or other platform image without a separate content review.
+- The reviewed `0.4.0` path covers linux/amd64 and linux/arm64. Do not activate
+  another platform image without a separate content review.
 - The image runs as the container's default root user and bundles the gateway
   source, package-manager tooling, native dependencies, and base-image
   SUID/SGID files. The registered command drops capabilities, prevents new
@@ -211,4 +234,4 @@ Agent:
 - [Unified AI System](https://github.com/happy520ai/unified-ai-system)
 - [60-second Codex MCP quickstart](https://github.com/happy520ai/unified-ai-system/blob/master/docs/codex-mcp-quickstart.md)
 - [MCP server guide](https://github.com/happy520ai/unified-ai-system/blob/master/packages/mcp-server/README.md)
-- [MCP image content review](https://github.com/happy520ai/unified-ai-system/blob/2967617bb266c79992a6ed40e70bb59e67f661ca/docs/security/mcp-image-review-0.3.2.md)
+- [MCP image content review](https://github.com/happy520ai/unified-ai-system/blob/4bbc5e81d1f372a5c80ba5597973f3284965adf6/docs/security/mcp-image-review-0.4.0.md)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-skills-collection",
-  "version": "4.0.21",
+  "version": "4.0.22",
   "description": "OpenCode CLI plugin that automatically downloads and keeps skills up to date.",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",