neomatrix369 · neomatrix369 · Aug 25, 2026 · Aug 25, 2026 · Aug 25, 2026 · Aug 25, 2026
@@ -8,6 +8,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Added
+- `/tw-verify` Quality column + blocked footer (slice 28): Scan Status table is
+  **Name | Type | Status | Quality | Note**; Quality shows Tessl
+  `items.quality_score` as **`N/100`** (else `—`); shared
+  **Will be blocked when Tripwire is enabled** appears once under the table;
+  **Sources** line attributes Quality to Tessl and security Status to
+  Cisco AI Defense and Snyk. Shared helpers in `guard/verify.py`; contract SSOT
+  [frontline-output-contract.md](docs/user-guide/frontline-output-contract.md).
+  Re-run `tripwire setup-agent-hooks` to refresh installed skills.
 - Docs: Wave **13-M** LLM usage / cost observability plan (slice 53) —
   [design/llm-usage-tracking.md](docs/design/llm-usage-tracking.md), slice stub,
   TRAIL/PROGRESS/STATUS DECIDED (not IMPLEMENTED). Historic collapsible Usage

@@ -19,7 +19,9 @@ from the repo directly — `tripwire setup-agent-hooks` installs copies:
   stdout; every failure mode is an explicit deny (fail closed).
 - **`hooks/_guard_entry.py`** — installed shim delegating to
   `guard.entry.main()` (run via `uv run --project <repo_root> --extra guard`).
-- **`skills/tw-verify`** — report scan status of named skills/MCP servers.
+- **`skills/tw-verify`** — report scan status of named skills/MCP servers
+  (RAG + Tessl Quality as `N/100` when present; blocked message as table footer;
+  **Sources** line: Quality = Tessl, Status = Cisco AI Defense + Snyk).
 - **`skills/tw-scan`** — submit named skills/MCP servers for scanning
   (`--force` or bare `force` to rescan valid artifacts).
 - **`skills/tw-enable` / `skills/tw-disable`** — flip the local `enable` key in

@@ -54,5 +54,5 @@ If `created` is true, the config file was missing and was recreated with default
 Tell the user, plainly:
 
 - Tripwire enforcement is now **OFF** — the Claude Code hook allows every skill/MCP call without checking scan status. Enforcement is **fully bypassed**, including red-rated and never-scanned artifacts.
-- Manual scanning and reporting still work while disabled: `/tw-scan` submits scans and `/tw-verify` reports status (its "will be blocked" column shows what enforcement WOULD do, marked as currently bypassed).
+- Manual scanning and reporting still work while disabled: `/tw-scan` submits scans and `/tw-verify` reports status (its blocked **footer** shows what enforcement WOULD do, marked as currently bypassed).
 - Re-enable at any time with `/tw-enable`.
@@ -58,15 +58,19 @@ for ident in sys.argv[1:]:
         blocked = not (rag == "green" or (rag == "amber" and threshold == "red"))
     else:
         blocked = True
+    item = s.get("item") or {}
+    q = item.get("quality_score")
+    quality_score = float(q) if isinstance(q, (int, float)) else None
     rows.append({
         "identifier": ident,
         "state": state,
         "rag": rag,
         "scanned_at": s.get("scanned_at"),
         "stale": state == "stale",
-        "errored": (s.get("item") or {}).get("heatmap_status") == "error",
+        "errored": item.get("heatmap_status") == "error",
         "changed": changed,
         "will_be_blocked": blocked,
+        "quality_score": quality_score,
     })
 print(json.dumps({
     "config": {
@@ -82,16 +86,18 @@ print(json.dumps({
 
 ## Step 4 — Report (Scan Status table only)
 
-Render exactly the tw-verify human table — same states, emoji, and labels (see `~/.claude/skills/tw-verify/SKILL.md` §Step 5). Parse the driver's JSON privately; do **not** dump a fenced `{config, artifacts}` block to the user. Summary of the row rules (N = `scan_validity_days`):
+Render exactly the tw-verify human table — same columns **Name | Type | Status | Quality | Note**, emoji/labels, Quality as **`N/100`** or `—`, blocked-message **footer**, and **Sources** line (Tessl Quality; Cisco/Snyk Status) (see `~/.claude/skills/tw-verify/SKILL.md` §Step 5 and [frontline-output-contract.md](../../../docs/user-guide/frontline-output-contract.md)). Parse the driver's JSON privately; do **not** dump a fenced `{config, artifacts}` block to the user. Summary of the row rules (N = `scan_validity_days`):
 
 - `fresh` + green → `🟢 GREEN (fresh)`, note `—`.
-- `fresh` + amber → `🟠 AMBER`; `Reported but not blocked at current threshold` when threshold is `red`, else bold **Will be blocked when Tripwire is enabled**.
-- `fresh` + red → `🔴 RED`; note ALWAYS bold **Will be blocked when Tripwire is enabled** — never omitted.
+- `fresh` + amber → `🟠 AMBER`; `Reported but not blocked at current threshold` when threshold is `red`, else distinct amber note + footer when blocked.
+- `fresh` + red → `🔴 RED`; note `rated red — at/above threshold` (blocked sentence is footer-only).
 - `stale` → `⚠️ STALE`; `Last scanned >N days ago — blocked until rescanned (run /tw-scan <name>)`.
 - `scanning` → `⏳ SCANNING`; `Scan in progress — check back shortly` (+ prior verdict if rag non-null).
-- `unscanned` → `🚫 UNSCANNED`; `Never scanned — blocked when Tripwire is enabled` (or `Last scan errored — resubmit (run /tw-scan <name>)` when errored).
-- `changed=true` → `✏️ CHANGED`; bold **content changed since last scan — run /tw-scan <name>** — takes precedence over every state-based row (the hook's tamper deny fires regardless of a green verdict), `will_be_blocked` true.
-- missing install dir → `❓ NOT FOUND`; bold **Will be blocked when Tripwire is enabled** — `Not installed under ~/.claude/skills — run tripwire setup-agent-hooks`.
+- `unscanned` → `🚫 UNSCANNED`; `Never scanned — offer /tw-scan <name>` (or `Last scan errored — resubmit (run /tw-scan <name>)` when errored).
+- `changed=true` → `✏️ CHANGED`; bold **content changed since last scan — run /tw-scan <name>** — takes precedence over every state-based row, `will_be_blocked` true.
+- missing install dir → `❓ NOT FOUND`; note `Not installed under ~/.claude/skills — run tripwire setup-agent-hooks` (footer covers blocked).
+
+If any `will_be_blocked`, print **Will be blocked when Tripwire is enabled** once under the table.
 
 If local `enable` is false, add the "currently bypassed" note; if `monitoring_enabled` is false, add the platform-switch warning.
 

@@ -5,7 +5,7 @@ description: Check the Tripwire scan status of Claude Code skills and MCP server
 
 # tw-verify
 
-Report the Tripwire scan status of one or more skills / MCP servers. Single pass over ALL requested names — never stop at the first problem; every requested name gets a row in the Scan Status table. Read-only: this skill never submits scans itself (it only offers to, at the end). Do **not** dump the driver's raw JSON (or any fenced `{config, artifacts}` block) to the user — that payload is for you to parse only.
+Report the Tripwire scan status of one or more skills / MCP servers. Single pass over ALL requested names — never stop at the first problem; every requested name gets a row in the Scan Status table. Columns follow the Frontline dual-output contract ([frontline-output-contract.md](../../../docs/user-guide/frontline-output-contract.md)): **Name | Type | Status | Quality | Note**, with Tessl Quality as **`N/100`** when `items.quality_score` is present (else `—`), the shared blocked phrase once as a **table footer**, and a **Sources** line (Quality = Tessl; Status = Cisco AI Defense + Snyk). Shared helpers: `guard.verify.verify_artifacts` / `format_quality_cell`. Read-only: this skill never submits scans itself (it only offers to, at the end). Do **not** dump the driver's raw JSON (or any fenced `{config, artifacts}` block) to the user — that payload is for you to parse only.
 
 **Hard rule — unscanned/blocked artifacts must not be executed.** If a row is `unscanned` (including errored), `stale`, `changed`, RED (or amber at threshold), or NOT FOUND: do **not** invoke that skill (`Skill` tool), do **not** call its `mcp__*` tools, and do **not** run its `install.sh` / scripts via Bash. Report that Tripwire will block those calls. Only offer `/tw-scan` via AskUserQuestion — never silently submit, and never “work around” a block by scanning so you can run the artifact in the same turn.
 
@@ -39,7 +39,7 @@ print(json.dumps({"resolutions": out}))
 
 Parse the JSON privately. For each entry:
 
-- `found=false` → ❓ NOT FOUND row (do not call the status driver for that name). Note: **Will be blocked when Tripwire is enabled** — no match in hook loci (`~/.claude/skills`, `.claude/skills`, `.mcp.json`, `~/.claude.json` incl. `projects["<cwd>"]`, `~/.tripwire/demo-mcp.json`, fixtures MCP manifest). Tip: demo skills are `safe-skill` / `vuln-skill` / `amber-skill`; demo MCP keys are `safe-tool` / `vuln-tool` / `amber-tool` (fixture names like `vuln-runtime-download` alias to those when demos are installed).
+- `found=false` → ❓ NOT FOUND row (do not call the status driver for that name). Note: no match in hook loci (`~/.claude/skills`, `.claude/skills`, `.mcp.json`, `~/.claude.json` incl. `projects["<cwd>"]`, `~/.tripwire/demo-mcp.json`, fixtures MCP manifest) — do **not** repeat the blocked phrase in the Note (it goes in the table footer). Tip: demo skills are `safe-skill` / `vuln-skill` / `amber-skill`; demo MCP keys are `safe-tool` / `vuln-tool` / `amber-tool` (fixture names like `vuln-runtime-download` alias to those when demos are installed).
 - `found=true` → use `identifier` (and `kind`) for Step 4. If `alias_of` is set, the Name column still shows the user’s requested name; optionally append `(as <resolved_as>)` in the Note.
 
 An absolute path the user passed that exists as a skill directory is resolved by the driver — do not invent your own path logic.
@@ -79,15 +79,19 @@ for ident in sys.argv[1:]:
         blocked = not (rag == "green" or (rag == "amber" and threshold == "red"))
     else:
         blocked = True
+    item = s.get("item") or {}
+    q = item.get("quality_score")
+    quality_score = float(q) if isinstance(q, (int, float)) else None
     rows.append({
         "identifier": ident,
         "state": state,
         "rag": rag,
         "scanned_at": s.get("scanned_at"),
         "stale": state == "stale",
-        "errored": (s.get("item") or {}).get("heatmap_status") == "error",
+        "errored": item.get("heatmap_status") == "error",
         "changed": changed,
         "will_be_blocked": blocked,
+        "quality_score": quality_score,
     })
 print(json.dumps({
     "config": {
@@ -101,40 +105,50 @@ print(json.dumps({
 ' <identifier-1> <identifier-2> ...
 ```
 
-The driver prints one JSON object `{config, artifacts}` — parse it privately; never paste it into the user-facing reply. NOT FOUND names never reach the driver (they get rows anyway). If the driver itself fails, report the failure and still render every row (status "unknown — status query failed"), never a partial silent result.
+The driver prints one JSON object `{config, artifacts}` — parse it privately; never paste it into the user-facing reply. Each artifact includes nullable `quality_score` (0–100 from `items.quality_score`); human render owns the `/100` display. NOT FOUND names never reach the driver (they get rows anyway, `quality_score` null). If the driver itself fails, report the failure and still render every row (status "unknown — status query failed"), never a partial silent result.
 
 ## Step 5 — Render the human-readable table
 
-One Markdown table, one row per REQUESTED name (selection rows count individually), all in a single pass. Use exactly these states, emoji, and labels (N = `scan_validity_days`):
+One Markdown table, one row per REQUESTED name (selection rows count individually), all in a single pass. Fixed columns: **Name | Type | Status | Quality | Note**. Use exactly these states, emoji, and labels (N = `scan_validity_days`). Quality cell: **`N/100`** when `quality_score` is a number for a skill (Tessl skill-review, 0–100, higher better); otherwise `—` (MCP, unscanned, scanning, not-found, null, or missing). Never a bare integer or `Q N` alone.
 
-| Name | Type | Status | Note |
-|------|------|--------|------|
-| `safe-skill` | skill | 🟢 GREEN (fresh) | — |
-| `vuln-skill` | skill | 🔴 RED | **Will be blocked when Tripwire is enabled** |
-| `safe-tool` | mcp | 🟠 AMBER | Reported but not blocked at current threshold |
-| `unknown-skill` | — | ❓ NOT FOUND | **Will be blocked when Tripwire is enabled** — no match in ~/.claude/skills, .claude/skills, .mcp.json, ~/.claude.json, ~/.tripwire/demo-mcp.json, fixtures manifest |
-| `old-skill` | skill | ⚠️ STALE | Last scanned >14 days ago — blocked until rescanned |
-| `pending-skill` | skill | ⏳ SCANNING | Scan in progress — check back shortly |
-| `new-skill` | skill | 🚫 UNSCANNED | Never scanned — blocked when Tripwire is enabled |
-| `edited-skill` | skill | ✏️ CHANGED | **content changed since last scan — run /tw-scan <name>** |
+| Name | Type | Status | Quality | Note |
+|------|------|--------|---------|------|
+| `safe-skill` | skill | 🟢 GREEN (fresh) | 91/100 | — |
+| `vuln-skill` | skill | 🔴 RED | 12/100 | rated red — at/above threshold |
+| `safe-tool` | mcp | 🟠 AMBER | — | Reported but not blocked at current threshold |
+| `unknown-skill` | — | ❓ NOT FOUND | — | no match in ~/.claude/skills, .claude/skills, .mcp.json, ~/.claude.json, ~/.tripwire/demo-mcp.json, fixtures manifest |
+| `old-skill` | skill | ⚠️ STALE | 80/100 | Last scanned >14 days ago — blocked until rescanned (run /tw-scan old-skill) |
+| `pending-skill` | skill | ⏳ SCANNING | — | Scan in progress — check back shortly |
+| `new-skill` | skill | 🚫 UNSCANNED | — | Never scanned — offer /tw-scan new-skill |
+| `edited-skill` | skill | ✏️ CHANGED | 70/100 | **content changed since last scan — run /tw-scan edited-skill** |
+
+**Blocked footer (de-dupe):** If **any** row has `will_be_blocked=true` (including NOT FOUND / RED / STALE / UNSCANNED / CHANGED / amber-at-threshold), print **once** under the table:
+
+**Will be blocked when Tripwire is enabled**
+
+Do **not** repeat that phrase in every Note. Row Notes keep *distinct* copy only.
+
+**Sources attribution:** Always print **once** under the table (after the blocked footer when present):
+
+*Sources: Quality is from Tessl (Review Quality). Security-related Status (GREEN/AMBER/RED) is from Cisco AI Defense and Snyk.*
 
 Row rules (map driver output → row):
 
 - `changed=true` → `✏️ CHANGED` — this row takes precedence over every state-based row (the hook's tamper deny fires regardless of a green verdict); note is the bold **content changed since last scan — run /tw-scan <name>**, and `will_be_blocked` is `true`.
 - `state=fresh, rag=green` → `🟢 GREEN (fresh)`, note `—`.
-- `state=fresh, rag=amber` → `🟠 AMBER`; note `Reported but not blocked at current threshold` when `threshold` is `red`, else **Will be blocked when Tripwire is enabled** (bold).
-- `state=fresh, rag=red` → `🔴 RED`; note is ALWAYS the bold **Will be blocked when Tripwire is enabled** — never omitted, no exceptions.
+- `state=fresh, rag=amber` → `🟠 AMBER`; note `Reported but not blocked at current threshold` when `threshold` is `red`, else note `amber at/above threshold` and `will_be_blocked=true` (footer covers the blocked sentence).
+- `state=fresh, rag=red` → `🔴 RED`; note `rated red — at/above threshold` (blocked sentence is footer-only).
 - `state=stale` → `⚠️ STALE`; note `Last scanned >N days ago — blocked until rescanned (run /tw-scan <name>)`.
 - `state=scanning` → `⏳ SCANNING`; note `Scan in progress — check back shortly`; if `rag` is non-null append `(prior verdict: <rag>)`.
-- `state=unscanned, errored=false` → `🚫 UNSCANNED`; note `Never scanned — blocked when Tripwire is enabled`.
+- `state=unscanned, errored=false` → `🚫 UNSCANNED`; note `Never scanned — offer /tw-scan <name>`.
 - `state=unscanned, errored=true` → `🚫 UNSCANNED`; note `Last scan errored — resubmit (run /tw-scan <name>)`.
-- unresolved name → `❓ NOT FOUND`; note ALWAYS the bold **Will be blocked when Tripwire is enabled** plus `— no match in ~/.claude/skills, .claude/skills, .mcp.json, ~/.claude.json, ~/.tripwire/demo-mcp.json, fixtures manifest`. Fail-closed: the hook denies any Skill/mcp__* call it cannot resolve to a known locus (same as unscanned).
+- unresolved name → `❓ NOT FOUND`; note `no match in ~/.claude/skills, .claude/skills, .mcp.json, ~/.claude.json, ~/.tripwire/demo-mcp.json, fixtures manifest`. Fail-closed: the hook denies any Skill/mcp__* call it cannot resolve to a known locus (same as unscanned); `will_be_blocked=true` → footer.
 
 The `run /tw-scan <name>` remedy in the STALE, errored, and CHANGED notes actually works because tw-scan always submits with `--force` — without force the CLI would skip unchanged content and a stale/errored state could never clear.
 
-After the table:
+After the table (and after the blocked footer when present):
 
-- If config `enabled` is `false`, add: `Note: Tripwire enforcement is currently DISABLED (/tw-disable) — "will be blocked" reports what enforcement would do when enabled; calls are currently bypassed.`
+- If config `enabled` is `false`, add: `Note: Tripwire enforcement is currently DISABLED (/tw-disable) — the blocked footer reports what enforcement would do when enabled; calls are currently bypassed.`
 - If `monitoring_enabled` is `false` while local `enable` is `true`, add a warning that the Supabase platform switch (`config.monitoring_enabled`) is OFF and still gates the guard — effective enforcement is local enable AND platform switch.
 
 ## Step 6 — Offer scans for blocked-but-fixable rows