fix(audit-hermes): handle fallback_providers as list or dict

The audit assumed fallback_providers was always a dict (single provider).
Two live agents (koby, koonimo) carry it as a LIST of dicts (one entry per
fallback), so the script crashed with:

    File "audit-hermes-config.py", line 211, in audit
        fb.get("provider") == "deepseek",
    AttributeError: 'list' object has no attribute 'get'

Both are REAL agent configs, so this is not a malformed-input case — the
script simply could not audit two of the four agents it exists to audit.

Fix:
- Normalize fallback_providers to a list of entries (dict → [dict], list → list)
- Apply the existing checks to each entry
- A malformed entry (not a mapping) produces a reported VIOLATION naming the
  offending entry, NOT an uncaught exception

Adds regression test using the real failing shape (list of dicts) and proves
it bites against the pre-fix revision.

Real audit results after fix:
- mumuni: FAIL — 7 violations
- tanko: FAIL — 21 violations
- koby: FAIL — 16 violations (previously crashed)
- koonimo: FAIL — 10 violations (previously crashed)

No agent configs were changed. No existing rules were relaxed.

audit-hermes-config.py

@@ -94,7 +94,16 @@ def audit(path):
         cfg = yaml.safe_load(f)
 
     model = cfg.get("model", {})
-    fb = cfg.get("fallback_providers", {})
+    fb_raw = cfg.get("fallback_providers", {})
+    # Normalize: fallback_providers may be a dict (single provider) or a list of dicts
+    # (one entry per fallback). Both shapes are valid; we must handle both without crashing.
+    if isinstance(fb_raw, dict):
+        fb_entries = [fb_raw]
+    elif isinstance(fb_raw, list):
+        fb_entries = fb_raw
+    else:
+        fb_entries = [fb_raw]  # Let it fail the check below as malformed
+    fb = fb_entries[0] if fb_entries else {}
     comp = cfg.get("compression", {})
     aux = cfg.get("auxiliary", {})
     deleg = cfg.get("delegation", {})
@@ -207,22 +216,32 @@ def audit(path):
         "Rule 14",
         f"delegation.provider must be 'harness' (got {deleg.get('provider')!r})",
     )
-    check(
-        fb.get("provider") == "deepseek",
-        "Rule 14",
-        f"fallback_providers.provider must be 'deepseek' (got {fb.get('provider')!r}) — "
-        f"true fallback diversity, not same endpoint as primary",
-    )
-    check(
-        fb.get("model") == "deepseek-v4-flash",
-        "Rule 14",
-        f"fallback_providers.model must be 'deepseek-v4-flash' (got {fb.get('model')!r})",
-    )
-    check(
-        fb.get("api_key_env") == "DEEPSEEK_API_KEY",
-        "Rule 14",
-        f"fallback_providers.api_key_env must be DEEPSEEK_API_KEY (got {fb.get('api_key_env')!r})",
-    )
+    # Check each fallback entry. A malformed entry (not a mapping) is a VIOLATION, not a crash.
+    for idx, entry in enumerate(fb_entries):
+        prefix = f"fallback_providers[{idx}]"
+        if not isinstance(entry, dict):
+            check(
+                False,
+                "Rule 14",
+                f"{prefix} must be a mapping (got {type(entry).__name__})",
+            )
+            continue
+        check(
+            entry.get("provider") == "deepseek",
+            "Rule 14",
+            f"{prefix}.provider must be 'deepseek' (got {entry.get('provider')!r}) — "
+            f"true fallback diversity, not same endpoint as primary",
+        )
+        check(
+            entry.get("model") == "deepseek-v4-flash",
+            "Rule 14",
+            f"{prefix}.model must be 'deepseek-v4-flash' (got {entry.get('model')!r})",
+        )
+        check(
+            entry.get("api_key_env") == "DEEPSEEK_API_KEY",
+            "Rule 14",
+            f"{prefix}.api_key_env must be DEEPSEEK_API_KEY (got {entry.get('api_key_env')!r})",
+        )
 
     # --- custom_providers sanity ---
     check(

contract-registry.yaml

@@ -1965,17 +1965,21 @@ contracts:
     verify_commands:
     - infisical run --env=prod -- python3 scripts/daily-infra-report.py --test-email
     - python3 -m pytest tests/test_daily_infra_report.py -q
-  email_dependency:
-    transport: smtp.gmail.com:587
-    identity: jtabiri@gmail.com
-    secret: EMAIL_PASSWORD (must be a Google app password)
-    status: DEGRADED as of 2026-09-25 - 534 5.7.9 Application-specific password required
-    note: A delivery failure is a credential dependency, not a code defect. Tracked
-      as daily-digest-mail-transport-20260921.
+  delivery:
+    transport: zulip-dm-attachment
+    recipient_user_id: 9
+    sender: abiba-bot@chat.sysloggh.net
+    key_source: abiba-bot Zulip key already on the execution host, read from the
+      600-mode env file /root/.pi/agent/extensions/zulip/.env
+    key_policy: do NOT add a vault entry - that is a captain decision under the auth-keys charter
+    body: short Markdown pointer; the HTML attachment IS the report
+    artifact: /var/log/daily-infra-report/infra-report-<UTCstamp>.html
+    note: Replaced SMTP/mail on 2026-09-26 by captain decision. Removes the Google
+      dependency entirely; closes daily-digest-mail-transport-20260921.
   exit_semantics:
-    '1': missing PVE_TOKEN, unreachable Proxmox probe, or failed email send - raises an alert
-    '0': healthy, or a deliberate DEGRADED leg where the email credential is absent
-      and the report is still produced
+    '1': missing PVE_TOKEN, unreachable Proxmox probe, missing/rejected Zulip
+      credential, or a failed upload/post - raises an alert
+    '0': healthy delivery only - there is no degraded delivery leg any more
   depends_on: []
   last_run: null
   last_status: null

daily-health-digest.prose.md

@@ -16,11 +16,12 @@ description: >
 
   Exit-code semantics (as they actually behave, verified 2026-09-25):
     * missing PVE_TOKEN, or an unreachable Proxmox probe -> exit 1 + alert
-    * missing EMAIL credential -> deliberate DEGRADED leg, exit 0, report still
-      produced
-    * email send failure -> exit 1 (a delivery fault, not a code defect)
+    * missing or rejected Zulip credential -> exit 1 (delivery is the only
+      output path, so it is a real failure, not a degraded leg)
+    * delivery failure -> exit 1, and the report body is printed AND persisted
+      so the content is never swallowed
 
-version: 1.0.0
+version: 2.0.0
 ---
 
 ## Purpose
@@ -93,14 +94,13 @@ $ infisical run --env=prod -- python3 scripts/daily-infra-report.py --json
 EXIT=0
 ```
 
-and in mail mode:
+and in delivery mode:
 
 ```
-   Sending email...
-   ✅ All legs fully credentialed
-📋 Summary:
-   Proxmox: 5/5 nodes online
-   VMs/CTs: 22/22 running
+   report ready: 16208 chars of HTML (delivered as a file attachment)
+   Sending to the captain's Zulip DM...
+   ✅ Delivered to Zulip DM (user 9), message id 86221, attachment 16208 bytes
+      at /user_uploads/2/45/m1cQesBFV78BGeNY2lN8xkN5/infra-report-20260926-153406.html
 ```
 
 Healthy means: every probe reports `ok`, `nodes_online == node_count`, and the
@@ -115,9 +115,8 @@ Verified on 2026-09-25 by running each case deliberately.
 | all probes reachable, email sent | 0 | — | healthy |
 | **missing `PVE_TOKEN`** | **1** | yes | `PROBE FAILURES: proxmox: node list unreachable (PVE_TOKEN missing or API down)`, and `cluster resources unreachable` |
 | **Proxmox probe unreachable** | **1** | yes | same path as above; `pve_probe_status: unreachable` |
-| **missing `EMAIL_PASSWORD`** | **0** | no | deliberate **DEGRADED** leg (`credential-missing: EMAIL_PASSWORD`); the report is still produced |
-| **email send fails** | **1** | yes | e.g. Gmail `534 5.7.9 Application-specific password required` |
-| degraded legs present (non-email) | 0 | no | logged under `⚠️ Degraded legs` |
+| **missing/rejected Zulip credential** | **1** | yes | delivery is the only output path; report printed and persisted |
+| **upload or message post fails** | **1** | yes | report printed and persisted; message names which step failed |
 
 The distinction is deliberate and must not be flattened:
 
@@ -130,22 +129,31 @@ The distinction is deliberate and must not be flattened:
 `PROBE_FAILURES` and `DEGRADED_LEGS` are separate lists for exactly this
 reason. Do not merge them.
 
-## Email-delivery dependency
+## Delivery: Zulip DM carrying the report as an HTML ATTACHMENT
 
-Delivery is a **credential dependency, not a code path**. The producer
-authenticates to `smtp.gmail.com:587` as `jtabiri@gmail.com` with
-`EMAIL_PASSWORD` from the vault and sends to `jerome@sysloggh.com`.
+Captain's decision 2026-09-26, clarified the same day: the digest is delivered to
+his **Zulip DM (user id 9)** from `abiba-bot@chat.sysloggh.net`, as an **HTML
+FILE** — an attachment, not HTML rendered in the message body and not a Markdown
+translation of it.
 
-* Since that Google account has two-step verification, `EMAIL_PASSWORD` must be
-  a Google **app password**, not the account password.
-* As of 2026-09-25 delivery is **failing** with
-  `534 5.7.9 Application-specific password required`; the fix is for the
-  captain to generate a fresh app password and place it in Infisical
-  (`infrastructure/production`) as `EMAIL_PASSWORD`.
-* **A delivery failure is not a code defect.** Investigation of a failed send
-  should start at the credential, not the script. Chasing it as a code bug
-  wastes the effort; verify the credential path first with `--test-email`.
-* Tracked separately as `daily-digest-mail-transport-20260921`.
+* the styled dashboard is built exactly as before and written to
+  `/var/log/daily-infra-report/infra-report-<UTCstamp>.html`;
+* it is uploaded through `POST /api/v1/user_uploads`;
+* the **message body stays short Markdown** — subject line, top-line status
+  (nodes online, guests running, any degraded legs), and a link to the
+  attachment. The attachment IS the report; the body does not reproduce it.
+
+This removes the Google dependency entirely: **no SMTP, no `EMAIL_PASSWORD`, no
+app password, nothing to rotate.** `daily-digest-mail-transport-20260921` is
+closed under this option.
+
+The **10,000-character message cap does not apply** — it bounds message TEXT
+only, and the report travels as a file. Do not shrink the report to fit it.
+
+The credential is abiba-bot's Zulip key already on the execution host at
+`/root/.pi/agent/extensions/zulip/.env` (`ABIBA_ZULIP_API_KEY`, mode 600,
+root-readable). **Do not place a new credential in the vault** — under the
+auth-keys charter that is a captain decision.
 
 ## What counts as a failure
 
@@ -153,10 +161,18 @@ A run FAILS (exit 1) when the report cannot be trusted or delivered:
 
 * any probe is unreachable, so a section would silently be empty;
 * `PVE_TOKEN` is missing;
-* the email send fails.
+* the Zulip credential is missing or rejected, or the upload/post fails.
 
-A run is DEGRADED (exit 0, report still produced) when a non-load-bearing
-credential is absent, currently only `EMAIL_PASSWORD`.
+There is **no degraded delivery leg any more**. Delivery is the only output
+path, so a missing credential is a failure rather than a survivable degradation —
+the previous "missing `EMAIL_PASSWORD` still exits 0" rule is retired with the
+mail transport.
+
+**A delivery failure must never swallow the report.** On failure the script
+prints the report body to stdout *and* leaves the HTML artifact on disk, so the
+content is always recoverable from the run log. That closes the queued defect
+where a failed send printed only the transport error and the report never
+surfaced.
 
 ## Failure behaviour
 
@@ -175,7 +191,7 @@ infisical run --env=prod -- python3 scripts/daily-infra-report.py --json \
   | grep -E 'pve_probe_status|node_count|nodes_online'
 
 # delivery path
-infisical run --env=prod -- python3 scripts/daily-infra-report.py --test-email
+infisical run --env=prod -- python3 scripts/daily-infra-report.py --test-zulip
 ```
 
 Regression tests: `tests/test_daily_infra_report.py` (7 tests). Four of them
@@ -183,5 +199,5 @@ fail against the pre-fix script, which is what makes them bite.
 
 ## Maintains
 
-- daily-infra-dashboard: { status: "degraded", reason: "email credential", last_check: timestamp }
+- daily-infra-dashboard: { status: "ok|undelivered", transport: zulip-dm-attachment, last_check: timestamp }
 - pve-probe: { status: "ok|unreachable", last_check: timestamp }

hermes-config-template.prose.md

@@ -5,6 +5,11 @@ description: >
   Standard Hermes configuration template for Syslog Solution LLC agents.
   Enforces shared infrastructure setup (Firecrawl, SearXNG, local models,
   RA-H OS MCP) while keeping agent-specific API keys and model choices.
+  UPDATED 2026-09-27: Clarified the Auxiliary Tasks policy — light aux (vision,
+  web_extract/browsing) -> gpu-vision (RTX 5070); context-heavy aux (compression) ->
+  syslog-auto (2026-07-23 decision, Rule 7). Removed the false "one model for all
+  auxiliary" / "never syslog-auto" claim; stated gpu-dense + strix-moe are the reasoning
+  hosts and aux should not be pinned to them. Now matches audit-hermes-config.py line-for-line.
   UPDATED 2026-08-07: Added litellm MCP server entry; updated Rule 15 (MCP Validation)
   to enforce REAL key headers (not env-vars) from the 2026-08-07 keyless-MCP incident.
   Added Rule 12 (Context-Issue Diagnostic) + Rule 13 (.env fallback enforcement) from the
@@ -167,13 +172,16 @@ compression:
   abort_on_summary_failure: false
 
 # ─── Auxiliary Tasks (CONSISTENCY RULE) ───
-# All auxiliary services MUST use identical model, base_url, and api_key_env:
-#   model: gpu-vision                # stable alias (NOT a raw model name)
+# Auxiliary tasks split into TWO model classes — do NOT assume one model for all:
+#   Light auxiliary (vision, web_extract/browsing) -> model: gpu-vision   # RTX 5070
+#     Keeps the reasoning hosts (gpu-dense / strix-moe) free for agent prompts.
+#   Context-heavy auxiliary (compression) -> model: syslog-auto           # weighted pool
+#     Deliberate per the 2026-07-23 OPERATIONAL DECISION in Rule 7: summarization
+#     runs against long histories and must be able to use the pool.
+# Do NOT pin auxiliary work to the reasoning hosts (gpu-dense / strix-moe).
+# All auxiliary services share identical ROUTING (base_url + api_key_env), not model:
 #   base_url: http://192.168.68.116/litellm/v1   # Rule 5 (2026-08-09): canonical authenticated; /v1 also OK
 #   api_key_env: LITELLM_API_KEY
-# Do NOT use syslog-auto for auxiliary tasks — it routes to the primary GPU.
-# gpu-vision = RTX 5070 (12B), freeing the Strix Halo for agent reasoning.
-# Heavy aux (delegation, x_search) use gpu-dense (RTX 3090) instead.
 # NEVER use retired model names (qwen3.6-27B-code, qwen3.6-35B-udq4; gemma-4-12b is retired
 # and no longer resolves) in agent configs — use the stable aliases so model swaps don't break agents.
 auxiliary:

memory-fixer.prose.md

@@ -6,13 +6,19 @@ name: memory-fixer
 description: >
   Auto-fix low-hanging fruit in the RA-H OS knowledge graph. No judgment calls — only deterministic Level 1 operations.
   Escalate anything that needs Kwame's input. Executes confirmed Kwame decisions to completion (state + updated_at).
-version: 2.1.0
+version: 2.2.0
 ---
 ---
 
 # Memory Fixer
 
-> **Canonical copy:** `/root/.hermes/contracts/memory-fixer-v3.md` (used by the `memory-fixer-daily` cron job). This file is the institutional record of the same contract. When the two diverge, treat the v3 source in `/root/.hermes/contracts/` as executable truth.
+> **Executable copy:** the `okyeame-memory-fixer` cron job on kagentz (`hermes cron list`) holds its instruction
+> set **inline in `~/.hermes/cron/jobs.json`** (`hermes cron edit <id> --prompt …`; there is no `--prompt-file`, and
+> `~/.hermes/cron/memory-fixer-prompt.md` is a synced draft, not the live instruction). This file is the institutional
+> record of the same contract; when the two diverge, the job prompt is what actually runs — diff it against this file
+> before claiming a prompt change landed.
+> ⚠️ Corrected 2026-09-26: the previous pointer (`/root/.hermes/contracts/memory-fixer-v3.md`) does not exist on
+> kagentz — no `/root` access from this container — and was verified unreachable, not merely stale.
 
 ## Purpose
 Auto-fix low-hanging fruit in the graph. No judgment calls — only deterministic Level 1 operations. Escalate anything that needs Kwame's input. When Kwame replies to an escalation, **execute the decision to completion** (update state and timestamps), never leaving a node in review-pending forever.
@@ -125,15 +131,44 @@ updateNode(id, {
 
 **Archive candidates are identified by the fix 3 query's `suggested_action = 'archive'` branch** (the `ELSE 'archive'` case: anything not an infrastructure/skill/documentation/strategic/audit type).
 
+### 5. Duplicate-Node Detection (Level 1 — read-only, every run)
+
+The graph's duplicate problem is rarely an agent mistyping a title: it is **recurring writers creating a new
+node per run instead of updating one**. This phase detects that class and reports it. It is read-only and
+**never merges**.
+
+```bash
+python3 /home/hermes/.hermes/scripts/memory_dup_detect.py --json
+```
+Read-only, ~15s over the whole graph, exit 0. That script is the source of truth for the clustering logic —
+do not re-implement it in the prompt or hand-count "duplicates" from titles.
+
+Consume each `items[]` entry's `verdict` field; do not invent your own:
+
+| `verdict` | Meaning | Required action |
+|---|---|---|
+| `WRITER-DEFECT` (`run_family: true`) | ONE scheduled task writes a new node per run | Report the ids, the `agents` (the writer) and `span_days`. **Never merge** — each node is that run's audit record. If the family grew since the last report, say `UNFIXED` and name the writer. |
+| `SAFE-MERGE` | Bodies identical | Still requires an explicit `merge #A into #B` decision from Kwame. |
+| `HUMAN-DECISION` | Same subject, bodies differ | Propose **connect (an edge)**, never merge. |
+
+- **Title overlap alone is not duplication.** Four distinct client workflows of one family (#357-#361) and two
+  different machines' migrations (#1792/#1793) both score high on title tokens while their bodies sit 0.1-0.3
+  apart. Confirm against body similarity before calling anything a duplicate.
+- Report clusters as **candidates for Kwame's decision**, never as established duplicates — a wrong auto-merge
+  destroys distinct content irrecoverably.
+- Per-run history nodes are kept deliberately. Bulk-merging a run family destroys the audit trail the family exists for.
+
 ## Level 2 Escalations (Kwame Decision Required)
 
 1. **Refresh-suggested stale nodes** flagged with `[REVIEW: refresh]` — refresh or keep? (Archive-suggested nodes are auto-archived under fix 4 and are not escalated.)
-2. **Duplicate Nodes** (same title or >70% title overlap) — Merge or keep?
+2. **Duplicate Nodes** — as detected by fix 5, by `verdict`, never by raw title overlap. `WRITER-DEFECT` is a writer fix (update one canonical node), not a merge decision; `SAFE-MERGE` and `HUMAN-DECISION` clusters are escalated for merge-or-connect.
 3. **Orphan Nodes >90 days old** — Archive or connect?
 
 ## Reporting Format
 
-The fixer reports to Kwame via this Zulip DM:
+The fixer does **not** send anything. Under the single-egress model (2026-09-21) every report leaves the node
+through Mumuni's gate (`comms_drop.py` for the queue, `comms_gate.py` to release and read-back verify), so
+exit 0 means QUEUED, never delivered. A report body is written to a file and handed to the outbox helper:
 
 ```
 🦅 Memory Fixer — [HH:MM UTC]
@@ -147,8 +182,10 @@ Stale nodes needing review (max 10):
 2. [Node #YYY] Title — Y days stale, SUGGEST: archive
 ...
 
-Duplicates needing decision:
-1. [Node #AAA] vs [Node #BBB] — Same title
+Duplicate clusters (candidates — Kwame decides; the fixer never merges unilaterally):
+1. [WRITER-DEFECT] #AAA/#BBB/#CCC — writer <agent>, N nodes, span Nd (UNFIXED if it grew since the last report)
+2. [HUMAN-DECISION] #DDD/#EEE — same subject, bodies differ, SUGGEST: connect
+3. "none" when the scan returned no clusters
 
 Orphans >90 days:
 1. [Node #EEE] Title — X days stale, orphaned
@@ -195,6 +232,7 @@ The result must be 0 rows when all decisions are executed. Report what was done.
 - **State integrity:** archived nodes have `state: archived` + `[ARCHIVED]` prefix; kept nodes are `state: active` without a `[REVIEW:]` tag.
 - **Auto-archive applied:** no node should ever be left tagged `[REVIEW: archive]` — that tag is retired. Any `[REVIEW: archive]` found means fix 4 was skipped; archive it and report.
 - **No review-pending forever:** after executing Kwame's decisions, `[REVIEW:%` node count must be 0.
+- **Duplicate scan ran:** every report carries the fix 5 block (`none` when there were no clusters). A report with no duplicate section means phase 5 was skipped — a silently skipped detection phase is the failure this phase exists to prevent.
 - **Timestamps:** every executed decision (and every auto-archive) bumps `updated_at`, so the node exits the stale window on the next run.
 
 ## Logging

scripts/daily-infra-report.py

@@ -243,7 +243,7 @@ def collect():
         ("Pulse", "https://pulse.sysloggh.net"),
         ("Proxmox", "https://192.168.68.12:8006"),
         ("SearXNG", "http://192.168.68.7:8888"),
-        ("Firecrawl", "http://192.168.68.7:3002/health"),
+        ("Firecrawl", "http://192.168.68.7:3002/"),  # Firecrawl serves no /health - the root is its liveness endpoint
     ]
     report["endpoints"] = []
     for name, url in endpoints:
@@ -389,6 +389,28 @@ def collect():
 
 # ── HTML Dashboard ──
 
+def classify_endpoint(code):
+    """Classify an endpoint probe per the fleet's probe policy.
+
+    Codified 2026-09-14 in the monitoring contracts: ANY HTTP status proves the
+    service answered, so the service is ALIVE - 200/301/302/401/403/404 alike.
+    Only a failed CONNECTION (000 / timeout / refused) is a failed probe. A 404
+    from a wrong path is not a service fault and must not render as one.
+
+    This replaces a string comparison that was wrong in both directions
+    (`ep["code"] >= "400"`): it rendered 301 as red, 404 as yellow, and a real
+    500 as yellow. 5xx is kept as its own "server error" signal rather than
+    being merged with 4xx.
+    """
+    if not code or code == "000":
+        return "red", "no connection"
+    if code.startswith("5"):
+        return "yellow", "server error"
+    if code.startswith(("2", "3", "4")):
+        return "green", "alive"
+    return "yellow", f"unexpected {code}"
+
+
 def build_html(r):
     issues = []
     
@@ -624,7 +646,7 @@ Proxmox: {r.get('pve_probe_status', 'ok')} ({r['nodes_online']}/{r['node_count']
     # ── Network Endpoints ──
     html += '<div class="card"><h2>🌐 Network Endpoints</h2><table><tr><th>Service</th><th>Status</th></tr>'
     for ep in r["endpoints"]:
-        color = "green" if ep["code"] in ("200","302","401") else ("yellow" if ep["code"] >= "400" else "red")
+        color = classify_endpoint(ep["code"])[0]
         html += f'<tr><td>{ep["name"]}</td><td class="{color}">HTTP {ep["code"]}</td></tr>'
     html += '</table></div>'
     
@@ -688,42 +710,152 @@ Proxmox: {r.get('pve_probe_status', 'ok')} ({r['nodes_online']}/{r['node_count']
     return html
 
 
-# ── Send Email ──
+# ── Delivery: Zulip DM carrying the report as an HTML ATTACHMENT ──
+#
+# Captain's decision, clarified 2026-09-26: the report is sent as an HTML FILE,
+# i.e. an attachment - NOT HTML rendered in the message body, and NOT a Markdown
+# translation of it. So the styled dashboard is built exactly as before, uploaded
+# through Zulip's file-upload API, and the message body stays short: subject,
+# top-line status, and a pointer to the attachment.
+#
+# This removes the Google dependency entirely (no SMTP, no EMAIL_PASSWORD).
+# The 10,000-character message cap does not apply: it bounds message TEXT only,
+# and the report travels as a file.
 
-def send_email(html_content, subject_prefix=""):
-    FROM = "abiba@sysloggh.com"
-    TO = "jerome@sysloggh.com"
-    SUBJECT = f"{subject_prefix}{'🏗️ Infrastructure Report — ' + DATE_STR}"
-    
-    msg = MIMEMultipart("alternative")
-    msg["From"] = FROM
-    msg["To"] = TO
-    msg["Subject"] = SUBJECT
-    msg.attach(MIMEText("Infrastructure report in HTML format — enable images to view.", "plain"))
-    msg.attach(MIMEText(html_content, "html"))
-    
+ZULIP_SITE = "https://chat.sysloggh.net"
+ZULIP_BOT_EMAIL = "abiba-bot@chat.sysloggh.net"
+CAPTAIN_USER_ID = 9
+ZULIP_KEY_FILE = "/root/.pi/agent/extensions/zulip/.env"
+REPORT_ARTIFACT_DIR = "/var/log/daily-infra-report"
+
+
+def zulip_key():
+    """abiba-bot's Zulip key, from the env or the on-host 600 file."""
+    key = os.environ.get("ABIBA_ZULIP_API_KEY")
+    if key:
+        return key.strip()
     try:
-        EMAIL_PASSWORD = os.environ.get("EMAIL_PASSWORD") or os.environ.get("SMTP_PASSWORD") or os.environ.get("MAIL_PASSWORD")
-        if not EMAIL_PASSWORD:
-            print("   ⚠️  Degraded leg: credential-missing: EMAIL_PASSWORD (or SMTP_PASSWORD/MAIL_PASSWORD)", file=sys.stderr)
-            DEGRADED_LEGS.append("credential-missing: EMAIL_PASSWORD")
-            return True, "✅ Email leg degraded (no credential) — report still produced"
-        GMAIL_EMAIL = "jtabiri@gmail.com"
-        
-        server = smtplib.SMTP("smtp.gmail.com", 587)
-        server.starttls()
-        server.login(GMAIL_EMAIL, EMAIL_PASSWORD)
-        server.sendmail(FROM, [TO], msg.as_string())
-        server.quit()
-        return True, "✅ Email sent to jerome@sysloggh.com"
-    except Exception as e:
-        return False, f"❌ Email failed: {e}"
+        with open(ZULIP_KEY_FILE) as fh:
+            for line in fh:
+                if line.startswith("ABIBA_ZULIP_API_KEY="):
+                    return line.split("=", 1)[1].strip()
+    except OSError:
+        return None
+    return None
+
+
+def build_summary(r, filename, test=False):
+    """Short Markdown body: subject, top-line status, pointer to the attachment.
+
+    Deliberately NOT a reproduction of the report - the attachment is the report.
+    """
+    nodes = f"{r.get('nodes_online', 0)}/{r.get('node_count', 0)} nodes online"
+    guests = f"{r.get('running_vms', 0)}/{r.get('total_vms', 0)} guests running"
+    lines = [
+        ("\U0001F9EA **TEST — **" if test else "") + "\U0001F3D7\uFE0F **Infrastructure Report — " + DATE_STR + "**",
+        f"**{nodes}** \u00b7 **{guests}** \u00b7 generated {TIME_STR}",
+    ]
+    problems = []
+    if r.get("pve_probe_status") != "ok":
+        problems.append(f"\u274c Proxmox probe: {r.get('pve_probe_status')}")
+    if r.get("resources_probe_status") != "ok":
+        problems.append(f"\u274c Resources probe: {r.get('resources_probe_status')}")
+    lit = r.get("litellm", {}) or {}
+    checks = lit.get("checks", []) or []
+    if checks:
+        passed = sum(1 for c in checks if c.get("status") == "pass")
+        if passed != len(checks):
+            problems.append(f"\u274c LiteLLM: {passed}/{len(checks)} checks pass")
+    if not (r.get("zulip_ext", {}) or {}).get("connected"):
+        problems.append("\u274c Zulip extension: not connected")
+    for leg in DEGRADED_LEGS:
+        problems.append(f"\u26a0\uFE0F degraded: {leg}")
+
+    lines.append("\n".join(problems) if problems else "\u2705 All monitored services healthy")
+    lines.append(f"\U0001F4CE **Full report attached:** `{filename}`")
+    return "\n\n".join(lines)
+
+
+def _curl(args, timeout=60):
+    r = subprocess.run(["curl", "-s", "-m", str(timeout)] + args,
+                       capture_output=True, text=True)
+    try:
+        return json.loads(r.stdout or "{}"), r.stdout
+    except json.JSONDecodeError:
+        return {}, r.stdout
+
+
+def _curl_json(args, timeout=90):
+    r = subprocess.run(["curl", "-s", "-m", str(timeout)] + args,
+                       capture_output=True, text=True)
+    try:
+        return json.loads(r.stdout or "{}"), r.stdout
+    except json.JSONDecodeError:
+        return {}, r.stdout
+
+
+def send_zulip(html_content, report, test=False):
+    """Upload the styled HTML and post a short pointer to the captain's DM.
+
+    Returns (ok, message). On ANY failure the report body is also printed to
+    stdout and persisted to disk, so a delivery failure can never swallow the
+    content - the defect this folds in.
+    """
+    os.makedirs(REPORT_ARTIFACT_DIR, exist_ok=True)
+    stamp = NOW.strftime("%Y%m%d-%H%M%S")
+    filename = f"infra-report-{stamp}.html"
+    html_path = os.path.join(REPORT_ARTIFACT_DIR, filename)
+    try:
+        with open(html_path, "w") as fh:
+            fh.write(html_content)
+    except OSError as e:
+        print(f"   \u26a0\uFE0F  could not persist report artifact: {e}", file=sys.stderr)
+
+    key = zulip_key()
+    if not key:
+        print(html_content)  # never swallow the content
+        return False, ("\u274c Delivery FAILED: no Zulip credential "
+                       "(ABIBA_ZULIP_API_KEY unset and "
+                       f"{ZULIP_KEY_FILE} unreadable). Report persisted to {html_path}")
+
+    auth = ["-u", f"{ZULIP_BOT_EMAIL}:{key}"]
+
+    # 1. Upload the report as a file.
+    up, up_raw = _curl_json(auth + [
+        "-X", "POST", f"{ZULIP_SITE}/api/v1/user_uploads",
+        "-F", f"file=@{html_path};type=text/html",
+    ])
+    if up.get("result") != "success" or not up.get("uri"):
+        print(html_content)
+        return False, (f"\u274c Delivery FAILED at upload: {up.get('msg') or up_raw[:160]} "
+                       f"(report persisted to {html_path})")
+
+    uri = up["uri"]
+    size = os.path.getsize(html_path)
+
+    # 2. Post a short message pointing at it.
+    body = build_summary(report, filename, test=test)
+    link = f"[{filename}]({uri})"
+    body = body.replace(f"`{filename}`", link)
+    payload, raw = _curl_json(auth + [
+        "-X", "POST", f"{ZULIP_SITE}/api/v1/messages",
+        "-d", "type=private",
+        "-d", f"to=[{CAPTAIN_USER_ID}]",
+        "--data-urlencode", f"content={body}",
+    ])
+    if payload.get("result") == "success":
+        return True, (f"\u2705 Delivered to Zulip DM (user {CAPTAIN_USER_ID}), "
+                      f"message id {payload.get('id')}, attachment {size} bytes at {uri}")
+
+    print(html_content)
+    return False, (f"\u274c Delivery FAILED at message post: {payload.get('msg') or raw[:160]} "
+                   f"(uploaded {uri}; report persisted to {html_path})")
 
 
 # ── Main ──
 
 if __name__ == "__main__":
-    is_test = "--test-email" in sys.argv
+    is_test = ("--test-email" in sys.argv) or ("--test-zulip" in sys.argv)
     
     print(f"{'🧪 TEST MODE' if is_test else '📊'} Collecting infrastructure data...")
     report = collect()
@@ -738,15 +870,14 @@ if __name__ == "__main__":
     
     print("   Building dashboard...")
     html = build_html(report)
-    
+    print(f"   report ready: {len(html)} chars of HTML (delivered as a file attachment)")
+
     if is_test:
-        prefix = "🧪 TEST — "
-        print("   Sending test email...")
+        print("   Sending TEST message to the captain's Zulip DM...")
     else:
-        prefix = ""
-        print("   Sending email...")
-    
-    ok, msg = send_email(html, subject_prefix=prefix)
+        print("   Sending to the captain's Zulip DM...")
+
+    ok, msg = send_zulip(html, report, test=is_test)
     print(f"   {msg}")
     
     # Show summary

tests/test_audit_hermes_fallback_list_shape.py

@@ -0,0 +1,236 @@
+"""Regression test for the fallback_providers list-shape crash in audit-hermes-config.py.
+
+WHY THIS FILE EXISTS: audit-hermes-config.py assumed `fallback_providers` was always a dict
+(single provider). Two live agents (koby, koonimo) carry it as a LIST of dicts (one entry per
+fallback), so the script crashed with:
+
+    File "audit-hermes-config.py", line 211, in audit
+        fb.get("provider") == "deepseek",
+    AttributeError: 'list' object has no attribute 'get'
+
+Both are REAL agent configs, so this is not a malformed-input case — the script simply could not
+audit two of the four agents it exists to audit. Until fixed, the key-hygiene check had no
+coverage for half the fleet while appearing to run.
+
+These tests execute the real CLI (`python3 audit-hermes-config.py <config>`) and assert:
+1. A config whose `fallback_providers` is a LIST of valid dicts does NOT crash (exit code is 0 or 1,
+   never a traceback/AttributeError).
+2. A config whose `fallback_providers` contains a MALFORMED entry (a list element that is not a
+   mapping) reports a VIOLATION naming the offending entry, NOT an uncaught exception.
+3. The dict shape still works (existing tests must stay green).
+
+No network, vault, or SSH access is required.
+"""
+from __future__ import annotations
+
+import pathlib
+import subprocess
+import sys
+
+ROOT = pathlib.Path(__file__).resolve().parent.parent
+AUDIT = ROOT / "audit-hermes-config.py"
+
+# A valid config where fallback_providers is a LIST of dicts (the real koby/koonimo shape).
+# One entry, well-formed: provider=deepseek, model=deepseek-v4-flash, api_key_env=DEEPSEEK_API_KEY.
+# This must produce a real verdict (PASS or FAIL) without crashing.
+LIST_SHAPE_VALID = """
+model:
+  api_key: ""
+  api_key_env: LITELLM_API_KEY
+  base_url: http://192.168.68.116/litellm/v1
+  max_tokens: 4096
+  default: syslog-auto
+  provider: harness
+fallback_providers:
+  - provider: deepseek
+    model: deepseek-v4-flash
+    api_key_env: DEEPSEEK_API_KEY
+compression:
+  model: syslog-auto
+  provider: harness
+  threshold: 0.65
+  max_context_window: 131072
+auxiliary:
+  vision:
+    model: gpu-vision
+    provider: harness
+  web_extract:
+    model: gpu-vision
+    provider: harness
+  compression:
+    model: syslog-auto
+    provider: harness
+delegation:
+  provider: harness
+custom_providers:
+  - name: harness
+    key_env: LITELLM_API_KEY
+    base_url: http://192.168.68.116/litellm/v1
+"""
+
+# A valid config where fallback_providers is a LIST with TWO entries (multiple fallbacks).
+# Both entries well-formed. Must not crash and should produce a real verdict.
+LIST_SHAPE_MULTI = """
+model:
+  api_key: ""
+  api_key_env: LITELLM_API_KEY
+  base_url: http://192.168.68.116/litellm/v1
+  max_tokens: 4096
+  default: syslog-auto
+  provider: harness
+fallback_providers:
+  - provider: deepseek
+    model: deepseek-v4-flash
+    api_key_env: DEEPSEEK_API_KEY
+  - provider: deepseek
+    model: deepseek-v4-flash
+    api_key_env: DEEPSEEK_API_KEY
+compression:
+  model: syslog-auto
+  provider: harness
+  threshold: 0.65
+  max_context_window: 131072
+auxiliary:
+  vision:
+    model: gpu-vision
+    provider: harness
+  web_extract:
+    model: gpu-vision
+    provider: harness
+  compression:
+    model: syslog-auto
+    provider: harness
+delegation:
+  provider: harness
+custom_providers:
+  - name: harness
+    key_env: LITELLM_API_KEY
+    base_url: http://192.168.68.116/litellm/v1
+"""
+
+# A config where fallback_providers is a LIST containing a MALFORMED entry:
+# one element is a plain string, not a mapping. The checker must report a VIOLATION
+# naming the offending entry (fallback_providers[1]) and NOT crash.
+LIST_SHAPE_MALFORMED = """
+model:
+  api_key: ""
+  api_key_env: LITELLM_API_KEY
+  base_url: http://192.168.68.116/litellm/v1
+  max_tokens: 4096
+  default: syslog-auto
+  provider: harness
+fallback_providers:
+  - provider: deepseek
+    model: deepseek-v4-flash
+    api_key_env: DEEPSEEK_API_KEY
+  - "not-a-mapping"
+compression:
+  model: syslog-auto
+  provider: harness
+  threshold: 0.65
+  max_context_window: 131072
+auxiliary:
+  vision:
+    model: gpu-vision
+    provider: harness
+  web_extract:
+    model: gpu-vision
+    provider: harness
+  compression:
+    model: syslog-auto
+    provider: harness
+delegation:
+  provider: harness
+custom_providers:
+  - name: harness
+    key_env: LITELLM_API_KEY
+    base_url: http://192.168.68.116/litellm/v1
+"""
+
+# The original DICT shape (single provider) must still work — existing behaviour preserved.
+DICT_SHAPE_VALID = """
+model:
+  api_key: ""
+  api_key_env: LITELLM_API_KEY
+  base_url: http://192.168.68.116/litellm/v1
+  max_tokens: 4096
+  default: syslog-auto
+  provider: harness
+fallback_providers:
+  provider: deepseek
+  model: deepseek-v4-flash
+  api_key_env: DEEPSEEK_API_KEY
+compression:
+  model: syslog-auto
+  provider: harness
+  threshold: 0.65
+  max_context_window: 131072
+auxiliary:
+  vision:
+    model: gpu-vision
+    provider: harness
+  web_extract:
+    model: gpu-vision
+    provider: harness
+  compression:
+    model: syslog-auto
+    provider: harness
+delegation:
+  provider: harness
+custom_providers:
+  - name: harness
+    key_env: LITELLM_API_KEY
+    base_url: http://192.168.68.116/litellm/v1
+"""
+
+
+def _run_config(tmp_path, name, text):
+    cfg = tmp_path / name
+    cfg.write_text(text)
+    proc = subprocess.run(
+        [sys.executable, str(AUDIT), str(cfg)],
+        capture_output=True, text=True,
+    )
+    return proc.returncode, proc.stdout, proc.stderr
+
+
+def test_list_shape_single_entry_does_not_crash(tmp_path):
+    """A LIST with one valid dict must not raise AttributeError; exit 0 (PASS)."""
+    code, out, err = _run_config(tmp_path, "list-single.yaml", LIST_SHAPE_VALID)
+    # Must NOT be a crash (traceback). A clean run exits 0 (PASS) or 1 (FAIL), never 2+ (exception).
+    assert code in (0, 1), f"Expected clean exit 0 or 1, got {code}\nSTDOUT:\n{out}\nSTDERR:\n{err}"
+    assert "AttributeError" not in err, f"Crashed with AttributeError:\n{err}"
+    assert "Traceback" not in err, f"Crashed with uncaught exception:\n{err}"
+    # The valid single-entry list should PASS (all rules satisfied).
+    assert code == 0, f"Expected PASS but got {code}\n{out}"
+    assert "RESULT: PASS" in out
+
+
+def test_list_shape_multiple_entries_does_not_crash(tmp_path):
+    """A LIST with two valid dicts must not raise AttributeError; exit 0 (PASS)."""
+    code, out, err = _run_config(tmp_path, "list-multi.yaml", LIST_SHAPE_MULTI)
+    assert code in (0, 1), f"Expected clean exit 0 or 1, got {code}\nSTDOUT:\n{out}\nSTDERR:\n{err}"
+    assert "AttributeError" not in err, f"Crashed with AttributeError:\n{err}"
+    assert "Traceback" not in err, f"Crashed with uncaught exception:\n{err}"
+    assert code == 0, f"Expected PASS but got {code}\n{out}"
+    assert "RESULT: PASS" in out
+
+
+def test_list_shape_malformed_entry_reports_violation_not_crash(tmp_path):
+    """A LIST containing a non-mapping element must be a reported VIOLATION, not a crash."""
+    code, out, err = _run_config(tmp_path, "list-malformed.yaml", LIST_SHAPE_MALFORMED)
+    # Must NOT be a crash.
+    assert "AttributeError" not in err, f"Crashed with AttributeError:\n{err}"
+    assert "Traceback" not in err, f"Crashed with uncaught exception:\n{err}"
+    # Should be a FAIL (exit 1) because the malformed entry is a violation.
+    assert code == 1, f"Expected FAIL (exit 1) but got {code}\n{out}"
+    assert "RESULT: FAIL" in out
+    # The violation must name the offending entry (fallback_providers[1]).
+    assert "fallback_providers[1]" in out, f"Violation did not name the offending entry:\n{out}"
+
+
+def test_dict_shape_still_passes(tmp_path):
+    """The original DICT shape (single provider) must still PASS — existing behaviour preserved."""
+    code, out, err = _run_config(tmp_path, "dict-valid.yaml", DICT_SHAPE_VALID)
+    assert code == 0, f"Expected PASS but got {code}\n{out}\nSTDERR:\n{err}"
+    assert "RESULT: PASS" in out