[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M72SWXPAJPVRZMRE0V80DJQ9":3,"skill-same:01M72SWXPAJPVRZMRE0V80DJQ9":49},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":28,"commitSha":29,"skillDirectory":30,"forksCount":14,"repositoryHtmlUrl":31,"repositoryHomepageUrl":31,"ownerHtmlUrl":32,"ownerBlogUrl":6,"ownerDescription":33,"updatedAt":34,"install":35,"resources":48},"01M72SWXPAJPVRZMRE0V80DJQ9","hermes-gateway-troubleshooting",null,"Debug Hermes Gateway connectivity and platform integration issues — Telegram, Discord, WhatsApp. Covers Python env corruption, systemd service config, plugin adapter changes, and credential\u002Fauth failures.","iberi22\u002Fskills","iberi22","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F10615454?v=4","01M72SQ2NXAA3GW0FSH053A5ES","Brahyan Belalcazar","01M72SQ2PYCH0SSA1K61NGEHDV",0,"MIT","https:\u002F\u002Fgithub.com\u002Fiberi22\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fhermes-gateway-troubleshooting\u002FSKILL.md",[],[],"2026-11-10 22:39:42",false,"---\nname: hermes-gateway-troubleshooting\ndescription: \"Debug Hermes Gateway connectivity and platform integration issues — Telegram, Discord, WhatsApp. Covers Python env corruption, systemd service config, plugin adapter changes, and credential\u002Fauth failures.\"\nversion: 1.0.0\nauthor: Hermes Agent\nplatforms: [linux]\n---\n\n# Hermes Gateway Troubleshooting\n\n## When to Use\n\nThis skill covers diagnosing and fixing issues with the **Hermes Gateway** — the systemd service that connects Hermes Agent to messaging platforms (Telegram, Discord, WhatsApp, etc.).\n\n**Triggers:**\n- \"Telegram no responde\" \u002F \"bot is not responding\"\n- Gateway starts but blocks messages\n- Gateway won't start \u002F crashes immediately\n- Platform integration stopped working after upgrade\n- \"ModuleNotFoundError\" or \"RuntimeError\" in gateway logs\n- Users are blocked as \"unauthorized\"\n\n**DO NOT use for:** CLI-only issues, Hermes TUI rendering. Model\u002Fprovider failures (\"The model provider failed after retries\") ARE covered here — see Layer 2.5.\n\n## Diagnostic Approach — Layer by Layer\n\nAlways work from outermost to innermost layer:\n\n1. **Is the process running?** → `systemctl --user status hermes-gateway.service`\n2. **Are there error logs?** → `tail -100 ~\u002F.hermes\u002Flogs\u002Fgateway.log`\n3. **Is the service config correct?** → `systemctl --user cat hermes-gateway.service`\n4. **Is the Python environment intact?** → Check uv tool site-packages\n5. **Is the plugin\u002Fadapter code correct?** → Check adapter's handling of config values\n\n---\n\n## Layer 1 — Process Health\n\n```bash\n# Check if gateway is running\nsystemctl --user status hermes-gateway.service\n\n# Check gateway PID and uptime\nps aux | grep \"gateway run\" | grep -v grep\n\n# Restart if needed\nsystemctl --user restart hermes-gateway.service\n```\n\n## Layer 2 — Log Analysis\n\n```bash\n# Full recent log\ntail -100 ~\u002F.hermes\u002Flogs\u002Fgateway.log\n\n# Only errors\ngrep -i \"error\\|traceback\\|exception\\|blocked\\|unauthorized\" ~\u002F.hermes\u002Flogs\u002Fgateway.log\n\n# Only Telegram activity\ngrep \"\\[Telegram\\]\" ~\u002F.hermes\u002Flogs\u002Fgateway.log\n\n# Follow in real time\ntail -f ~\u002F.hermes\u002Flogs\u002Fgateway.log\n```\n\n**Key log signatures:**\n\n| Log line | Meaning | Likely Fix |\n|----------|---------|------------|\n| `Blocked unauthorized user X` | Allowlist check failed | Check `allow_from` format in config.yaml |\n| `ModuleNotFoundError: No module named 'pydantic_core._pydantic_core'` | Python ABI mismatch | Check uv tool has correct site-packages |\n| `Failed to initialize OpenAI client` | Python deps broken | Reinstall missing packages |\n| `Connected to Telegram (polling mode)` | OK | — |\n| `set_my_commands OK` | OK | — |\n| `ImportError: cannot import name '\u003CX>' from '\u003Cmodule>'` | Version skew — daemon running pre-update code against post-update files | See Layer 1.5 |\n\n## Layer 2.4 — Version Skew: gateway process predating a code update\n\n**Symptom:** every inbound message fails with `ImportError: cannot import name\n'X' from 'Y'`, naming a symbol that plainly exists in the file on disk. Multiple\nunrelated symbols fail at once, from modules that import fine interactively.\n\n**Cause:** `hermes update` (or a git checkout of a new tag) replaces the source\ntree while the long-running gateway keeps the OLD modules already imported in\nmemory. The process holds a mix of old code and new files: old call sites\nimport symbols the new layout moved or renamed. Nothing on disk is broken.\n\n**Diagnose — compare process start time against the update, not against the file:**\n\n```bash\n# When did the running gateway start?\nps -o pid,lstart,etime -p \"$(pgrep -f 'hermes_cli.main gateway run' | head -1)\"\n\n# When was the source last replaced?\nstat -c '%y' ~\u002F.hermes\u002Fhermes-agent\u002Fhermes_cli\u002Fproviders.py\ngit -C ~\u002F.hermes\u002Fhermes-agent log -1 --format='%ci %s'\n```\n\nIf the process started BEFORE the update, version skew is the diagnosis.\n\n**Confirm the disk side is actually healthy** — this is the step that separates\nskew from a real breakage:\n\n```bash\n~\u002F.hermes\u002Fhermes-agent\u002Fvenv\u002Fbin\u002Fpython -c \"from hermes_cli.providers import LLAMACPP_ALIASES; print('imports OK')\"\n~\u002F.hermes\u002Fhermes-agent\u002Fvenv\u002Fbin\u002Fpython -c \"from tools.mcp_tool_common import mcp_server_enabled; print('imports OK')\"\n```\n\nIf those succeed, the fix is a restart, not a code change or reinstall:\n\n```bash\nsystemctl --user restart hermes-gateway.service\n```\n\n**Verify by scoping errors to the new process, never by counting the whole log.**\n`grep -c ImportError gateway.log` is cumulative and will show the historical\nfailures forever:\n\n```bash\nawk '\u002F\u003Crestart timestamp>\u002F,0' ~\u002F.hermes\u002Flogs\u002Fgateway.log | grep -c ImportError   # must be 0\ngrep -E \"telegram connected\" ~\u002F.hermes\u002Flogs\u002Fgateway.log | tail -1\n```\n\nHealthy reconnect looks like:\n```\n[Telegram] Telegram polling confirmed healthy: getUpdates progressing (generation 1)\ngateway.run: ✓ telegram connected\n```\n\n**PITFALL — an update procedure that does not restart the gateway leaves the\nsystem silently broken.** Any task that replaces Hermes source (update, tag\ncheckout, dependency sync) MUST end with the gateway restart plus the\npost-restart connectivity check. Report the update as incomplete until\n`✓ telegram connected` appears with a process start time newer than the update.\n`hermes doctor` does NOT catch this — it inspects on-disk config, not the\nrunning process's loaded modules.\n\n## Layer 2.5 — Model\u002FProvider Failure Diagnosis\n\nWhen a session shows `⚠️ The model provider failed after retries. I kept raw provider details out of chat; check gateway logs for diagnostics.` — the raw details ARE in the logs, the chat message just hides them by design:\n\n```bash\n# errors.log (NOT gateway.log) carries per-attempt provider details\ngrep -i \"api call failed\" ~\u002F.hermes\u002Flogs\u002Ferrors.log | tail -20\ngrep -i \"API call failed\" ~\u002F.hermes\u002Flogs\u002Ferrors.log | grep \"$(date +%F)\" | tail -20   # solo hoy\n```\n\nEach failed attempt logs `error_type=...` + `summary=...` for the exact provider\u002Fmodel. Taxonomy:\n\n| error_type | summary signal | Meaning | Action |\n|---|---|---|---|\n| `APIStatusError` | `HTTP 402 ... Insufficient Balance` | **Provider account balance exhausted (user-actionable)** | Top up the provider account \u002F swap API key |\n| `InternalServerError` | `HTTP 500: {\"type\":\"Router.Unavailable\",\"modelID\":...}` | Provider router can't serve that model right now (transient) | Retry \u002F wait \u002F temporarily switch model in config.yaml |\n| `InternalServerError` | `HTTP 500: Internal server error` | Provider-side outage | Retry with backoff (Hermes already retries 3x) |\n| `APIConnectionError` | `Connection error.` | Provider unreachable (network\u002Foutage) | Retry; `curl -sI \u003Cbase_url>`; if ALL providers fail it's network |\n\nWorkflow:\n1. Get the timestamp of the failed turn from `agent.log`: `grep \"model provider failed\" ~\u002F.hermes\u002Flogs\u002Fagent.log* | tail -5`\n2. Grep `errors.log` around that timestamp for `API call failed` → read `error_type` + `summary`\n3. Classify: **user-actionable** (402 balance, 401 bad key — the user must act) vs **transient** (500\u002Fconnection — retry works, no user action)\n4. Check whether the SAME error persists now (`tail errors.log`) — a one-off 500 vs a persistent `Router.Unavailable` are different answers\n5. Also watch `agent.credential_pool: no available entries (all exhausted or empty)` — credential pool empty compounds the failure\n\nReal case 2026-08-02\u002F03: chat error caused first by `HTTP 402 Insufficient Balance` (opencode.ai account empty — user needed to top up), then the next day by `HTTP 500 Router.Unavailable modelID=deepseek-v4-flash` (provider-side, transient). Two different causes on consecutive days, both visible ONLY in errors.log.\n\n## Layer 2.6 — Hermes Execution Health Audit (logs + entorno, verificado 2026-08-05)\n\nCuando el usuario pide \"revisa los logs y todo alrededor para mejorar la ejecución de Hermes\", la\nauditoría NO es solo del gateway — es de todo el proceso. Pasos que producen hallazgos accionables:\n\n```bash\n# 1. Tamaño y rotación de logs — agent.log rota a ~5MB; errors.log a ~2MB\nls -la ~\u002F.hermes\u002Flogs\u002F\n# 2. Tipos de error en agent.log (top tool failures del turno actual)\ngrep \"WARNING\" ~\u002F.hermes\u002Flogs\u002Fagent.log | grep -oE \"Tool [a-z_]+ returned error|Tool [a-z_]+: |agent\\.[a-z_]+:\" | sort | uniq -c | sort -rn | head -8\n# 3. Caídas del gateway — gateway-exit-diag.log registra el traceback del exit code\ntail -20 ~\u002F.hermes\u002Flogs\u002Fgateway-exit-diag.log\njournalctl --user -u hermes-gateway.service --since \"3 days ago\" | grep -cE \"Started|Stopped\"   # reinicios\n# 4. Recursos: RAM del gateway y pico histórico (pico 28GB en esta máquina = posible OOM previo)\nsystemctl --user status hermes-gateway.service | grep -E \"Active|Memory|Tasks\"\n# 5. state.db creciendo (sesiones\u002Fmensajes almacenados — 2.6GB\u002F245K msgs en esta máquina)\nls -lh ~\u002F.hermes\u002Fstate.db\n```\n\n**Hallazgo recurrente — background curator ruidoso (34+ errores\u002Fturno):** el self-improvement\nreview corre después de cada turno e intenta parchear skills vía `skill_manage`, pero se le NEGA\nsistemáticamente, llenando errors.log:\n```\nRefusing background curator patch for skill 'X': the skill records show it is not agent-created\n(created_by=None). Manually authored skills are off-limits... \u002F current SKILL.md content has not\nbeen loaded in this review turn...\n```\n- **Causa**: skills escritas a mano (created_by=None en `.usage.json`) o el review no carga el\n  SKILL.md antes de parchear → el guard las rechaza. No es un fallo del skill, es el review\n  intentando auto-mejorar contenido que no le corresponde.\n- **Impacto**: llamadas de API fallidas por turno + logs inundados. Si `errors.log` está lleno de\n  este patrón, es la fuente de ruido #1.\n- **Fix propuesto**: configurar `auxiliary.background_review` a un modelo barato explícito o\n  desactivar el auto-parchado de skills manuales (dejar solo captura de memoria).\n- **PITFALL de patch a skills manuales**: el mismo guard bloquea `skill_manage patch\u002Fedit` sobre\n  skills con `created_by=None` (ej. skills creadas antes de que el registro `.usage.json`\n  existiera, o cuyo create inicial falló y dejó el registro huérfano). No es un error de la\n  operación — es el guard anti-curation. Workaround: si el skill está protegido pero el\n  aprendizaje es importante, documentarlo en un skill editable (los que tienen\n  `author: Hermes Agent` en frontmatter) o crear uno nuevo.\n\n**gateway exit code 75 (verificado varias veces 2026-08-01..03):** `SystemExit(75)` en\n`gateway\u002Frun.py:22936` — el gateway se cae con este código cuando el runner aborta. Diagnóstico\nen `gateway-exit-diag.log` (traceback completo con PID + timestamp). Los reinicios frecuentes\n(7 en 3 días) con pico de RAM de 28GB apuntan a OOM por delegaciones masivas o compresión de\ncontexto — vigilar memoria antes de asumir problema de código.\n\n**state.db 2.6GB (245K mensajes, 3.4K sesiones):** cada `session_search` indexa todo el DB;\nbackups lentos. Mantenimiento: podar sesiones >60 días (exportar a Xavier antes) + `VACUUM`.\n\n## Layer 1.5 — Version Skew After an In-Place Update\n\nThe highest-yield check when a gateway breaks right after `hermes update`: the\nrunning process holds the OLD code in memory while the files on disk are NEW.\nEvery inbound message then dies on an import that no longer resolves.\n\nSignature in `~\u002F.hermes\u002Flogs\u002Fgateway.log`:\n\n```\nImportError: cannot import name 'LLAMACPP_ALIASES' from 'hermes_cli.providers'\nImportError: cannot import name 'mcp_server_enabled' from 'tools.mcp_tool_common'\n```\n\nRead as a module-move in the new release, not as a broken install.\n\n**Triage — compare the process start time against the file mtime:**\n\n```bash\nps -o pid,lstart,etime -p $(pgrep -f \"hermes_cli.main gateway run\" | head -1)\nstat -c '%y' ~\u002F.hermes\u002Fhermes-agent\u002Fhermes_cli\u002Fproviders.py\n```\n\nProcess started hours BEFORE the update ⇒ skew. Confirm the symbol is fine with\ncurrent code before touching anything:\n\n```bash\n~\u002F.hermes\u002Fhermes-agent\u002Fvenv\u002Fbin\u002Fpython -c \\\n  \"from hermes_cli.providers import LLAMACPP_ALIASES; print('imports OK')\"\n```\n\nIf that succeeds, the fix is a restart and nothing else:\n\n```bash\nsystemctl --user restart hermes-gateway.service\n```\n\nThen verify the NEW process, and prove the skew is gone:\n\n```bash\nps -o lstart= -p $(pgrep -f \"hermes_cli.main gateway run\" | head -1)   # fresh\ngrep -a \"✓ telegram connected\" ~\u002F.hermes\u002Flogs\u002Fgateway.log | tail -1\n# count errors only AFTER the new start time — a whole-file count stays non-zero forever\nawk '\u002F\u003Cnew start timestamp>\u002F,0' ~\u002F.hermes\u002Flogs\u002Fgateway.log | grep -c ImportError   # expect 0\n```\n\n**PITFALL:** `grep -c ImportError gateway.log` is cumulative and never returns 0\nafter an update. Always scope the count to the region after the restart, or you\nwill report a resolved fault as a live one.\n\n**PITFALL:** `gateway.log` may be detected as a binary file, making `grep` print\n`coincidencia en fichero binario` and silently match nothing. Add `-a` (as\nabove) whenever you grep a log that has grown large or contains non-UTF8 bytes.\n\n**Prevent it:** stop the gateway before updating, or restart it immediately\nafter, as one atomic task. An update that leaves a long-lived daemon running\nagainst replaced source will break every platform adapter until that restart.\n\n## Layer 3 — Systemd Service Config\n\n```bash\n# View the full service unit\nsystemctl --user cat hermes-gateway.service\n\n# Check for these critical fields:\n# - WorkingDirectory: must point to CURRENT python version's site-packages\n# - Environment=PATH: must reference correct python version\n# - ExecStart: correct python binary path\n```\n\n### Common Service File Issues\n\n**WorkingDirectory pointing to old Python version:**\n```\n# WRONG (stale python3.13):\nWorkingDirectory=\u002Fhome\u002Fuser\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Flib\u002Fpython3.13\u002Fsite-packages\n\n# RIGHT (current python3.14):\nWorkingDirectory=\u002Fhome\u002Fuser\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Flib\u002Fpython3.14\u002Fsite-packages\n```\n\n### Fix\n```bash\n# Edit the service file\nnano ~\u002F.config\u002Fsystemd\u002Fuser\u002Fhermes-gateway.service\n# Then:\nsystemctl --user daemon-reload\nsystemctl --user restart hermes-gateway.service\n```\n\n## Layer 4 — Python Environment (uv Tool)\n\n### Detect Dual-Version Site-Packages\n\nWhen uv upgrades the hermetic Python inside a tool, old version-specific site-packages can linger and cause import shadowing (Python resolves modules from the stale directory first).\n\n```bash\n# Check which python versions exist in the uv tool\nls ~\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Flib\u002F\n\n# Check actual python version of the tool's binary\n~\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Fbin\u002Fpython3 --version\n\n# If you see python3.13\u002F AND python3.14\u002F, the old one is stale\n# Fix: ensure systemd WorkingDirectory points to the CURRENT version\n# Fix: install missing packages into the current version's site-packages\n```\n\n**Root cause:** Python adds the current working directory to `sys.path` as `''` (empty string). If `WorkingDirectory` points to an old pythonX.Y\u002Fsite-packages, Python finds packages there first — even when `sys.path` doesn't explicitly include that directory.\n\n### Install Missing Packages\n\nWhen the correct site-packages directory lacks packages that only exist in a stale version:\n\n```bash\nuv pip install --python ~\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Fbin\u002Fpython3 \\\n  pydantic pydantic-core \\\n  python-telegram-bot \\\n  anthropic openai \\\n  rich tenacity httpx[socks] edge-tts \\\n  certifi python-dotenv fire pyyaml \\\n  jinja2 psutil tabulate termcolor wcwidth\n```\n\n### Verify\n```bash\n~\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Fbin\u002Fpython3 -c \"\nimport pydantic_core\nimport pydantic\nimport telegram\nimport openai\nprint('pydantic_core:', pydantic_core.__version__)\nprint('pydantic:', pydantic.__version__)\nprint('ALL OK')\n\"\n```\n\n## Layer 5 — Plugin\u002FAdapter Config\n\n### Telegram `allow_from` Format\n\nHermes 0.19.0+ uses a plugin-based Telegram adapter (`plugins\u002Fplatforms\u002Ftelegram\u002F`). The `allow_from` key is read from `config.yaml`'s `telegram:` section and bridged into the plugin's `extra` dict via `gateway\u002Fconfig.py` (~line 1472-1473 → `extra.update(bridged)` at ~line 1524).\n\n**Critical:** The adapter iterates over `allow_from` to build an allowed set. Four formats can arrive:\n\n| YAML config | Python type | Iteration behavior |\n|---|---|---|\n| `allow_from: [2076598024]` | `list[int]` | ✅ One iteration per ID |\n| `allow_from:\\n  - '2076598024'` | `list[str]` | ✅ One iteration per ID |\n| `allow_from: 2076598024` | `int` | ❌ `TypeError: 'int' object is not iterable` (traceback at `adapter.py:1019`) |\n| `allow_from: '2076598024'` | `str` | ❌ Iterates character-by-character (`'2'`, `'0'`, ...) |\n| `allow_from: '[\"2076598024\"]'` | `str` | ❌ Produces literal set `{'[\"2076598024\"]'}` — never matches |\n\nThe last row is the **`hermes config set` pitfall**: `hermes config set telegram.allow_from '[id]'` stores the value as a **literal quoted string** `'[\"id\"]'` in YAML, **NOT** as a native YAML list. Always edit `config.yaml` directly or use `sed`\u002FPython to write the YAML list format properly.\n\n**Preferred config format:**\n```yaml\ntelegram:\n  allow_from:\n    - '2076598024'       # YAML list with one quoted string per user ID\n```\n\nMultiple users:\n```yaml\ntelegram:\n  allow_from:\n    - '2076598024'\n    - '1234567890'\n```\n\n**Adapter patch** (shown with line numbers from `adapter.py:1016-1020`):\n\nThe original code assumes `allow_from` is always a list:\n```python\n# Original (broken for int\u002Fstr):\nadapter_allow_from = self.config.extra.get(\"allow_from\")\nif adapter_allow_from is not None:\n    allowed = {str(u).strip() for u in adapter_allow_from if str(u).strip()}\n    return user_id in allowed or \"*\" in allowed\n```\n\nDefensive fix — wrap non-list values before iterating:\n```python\nadapter_allow_from = self.config.extra.get(\"allow_from\")\nif adapter_allow_from is not None:\n    if isinstance(adapter_allow_from, str):\n        adapter_allow_from = [adapter_allow_from]       # str → list\n    elif not isinstance(adapter_allow_from, (list, tuple)):\n        adapter_allow_from = [str(adapter_allow_from)]  # int\u002Ffloat → list\n    allowed = {str(u).strip() for u in adapter_allow_from if str(u).strip()}\n    return str(user_id) in allowed or \"*\" in allowed\n```\n\n**Key details:**\n- Use `str(user_id)` for comparison — Telegram may pass `user_id` as int while the allowed set contains strings\n- After patching `adapter.py`, **restart the gateway** via `systemctl --user restart hermes-gateway.service` — the code change is hot only after restart\n- The config → plugin bridging happens in `gateway\u002Fconfig.py` during gateway startup; changes to `config.yaml` only take effect after restart\n- If you still get errors after fixing config, check that `systemctl --user daemon-reload` ran if you edited the service file\n\n### Config → Plugin Bridging\n\nThe gateway config loader (`gateway\u002Fconfig.py`) reads top-level platform blocks from `config.yaml` and bridges specific keys into the plugin's `extra` dict. Bridged keys include:\n\n`allow_from`, `allowed_chats`, `group_allowed_chats`, `allowed_topics`, `require_mention`, `dm_policy`, `group_policy`, `gateway_restart_notification`, `typing_indicator`, `channel_prompts`, `free_response_channels`, `mention_patterns`\n\n## Full Debugging Sequence\n\nWhen Telegram (or any platform) stops responding:\n\n```\n1. systemctl --user status hermes-gateway.service\n   → Is it running? If not, start it.\n\n2. tail -50 ~\u002F.hermes\u002Flogs\u002Fgateway.log | grep -E \"ERROR|Traceback|Blocked|unauthorized\"\n   → What errors exist?\n\n3. If `TypeError: 'int' object is not iterable` at `adapter.py:1019`:\n   → `allow_from` in config.yaml is an unquoted integer parsed by YAML as `int`\n   → Fix: change to proper YAML list format\n   → **WARNING:** `hermes config set telegram.allow_from '[id]'` stores a JSON-stringified\n     literal string `'[\"id\"]'`, NOT a YAML list. Edit config.yaml directly.\n   → If the adapter doesn't handle all formats (str\u002Fint\u002Flist), patch adapter.py\n   → **Always restart gateway after adapter patch** (`systemctl --user restart hermes-gateway.service`)\n\n4. If \"Blocked unauthorized user YOUR_ID\":\n   → Check telegram.allow_from format in config.yaml (same fix as above)\n   → Ensure the user_id in config matches the Telegram user ID exactly\n\n5. If pydantic_core ModuleNotFoundError:\n   → Check uv tool python version vs site-packages ABI\n   → Install missing packages for current python version\n\n6. If code loads from wrong site-packages:\n   → Check systemd service WorkingDirectory\n   → Update to point to current python version\n\n7. Restart gateway and test.\n```\n\n## Verification\n\n```bash\n# Send a test message via Telegram and check logs\ngrep -E \"inbound message|Processed message|Agent error\" ~\u002F.hermes\u002Flogs\u002Fgateway.log | tail -5\n```\n","# Hermes Gateway Troubleshooting\n\n## When to Use\n\nThis skill covers diagnosing and fixing issues with the **Hermes Gateway** — the systemd service that connects Hermes Agent to messaging platforms (Telegram, Discord, WhatsApp, etc.).\n\n**Triggers:**\n- \"Telegram no responde\" \u002F \"bot is not responding\"\n- Gateway starts but blocks messages\n- Gateway won't start \u002F crashes immediately\n- Platform integration stopped working after upgrade\n- \"ModuleNotFoundError\" or \"RuntimeError\" in gateway logs\n- Users are blocked as \"unauthorized\"\n\n**DO NOT use for:** CLI-only issues, Hermes TUI rendering. Model\u002Fprovider failures (\"The model provider failed after retries\") ARE covered here — see Layer 2.5.\n\n## Diagnostic Approach — Layer by Layer\n\nAlways work from outermost to innermost layer:\n\n1. **Is the process running?** → `systemctl --user status hermes-gateway.service`\n2. **Are there error logs?** → `tail -100 ~\u002F.hermes\u002Flogs\u002Fgateway.log`\n3. **Is the service config correct?** → `systemctl --user cat hermes-gateway.service`\n4. **Is the Python environment intact?** → Check uv tool site-packages\n5. **Is the plugin\u002Fadapter code correct?** → Check adapter's handling of config values\n\n---\n\n## Layer 1 — Process Health\n\n```bash\n# Check if gateway is running\nsystemctl --user status hermes-gateway.service\n\n# Check gateway PID and uptime\nps aux | grep \"gateway run\" | grep -v grep\n\n# Restart if needed\nsystemctl --user restart hermes-gateway.service\n```\n\n## Layer 2 — Log Analysis\n\n```bash\n# Full recent log\ntail -100 ~\u002F.hermes\u002Flogs\u002Fgateway.log\n\n# Only errors\ngrep -i \"error\\|traceback\\|exception\\|blocked\\|unauthorized\" ~\u002F.hermes\u002Flogs\u002Fgateway.log\n\n# Only Telegram activity\ngrep \"\\[Telegram\\]\" ~\u002F.hermes\u002Flogs\u002Fgateway.log\n\n# Follow in real time\ntail -f ~\u002F.hermes\u002Flogs\u002Fgateway.log\n```\n\n**Key log signatures:**\n\n| Log line | Meaning | Likely Fix |\n|----------|---------|------------|\n| `Blocked unauthorized user X` | Allowlist check failed | Check `allow_from` format in config.yaml |\n| `ModuleNotFoundError: No module named 'pydantic_core._pydantic_core'` | Python ABI mismatch | Check uv tool has correct site-packages |\n| `Failed to initialize OpenAI client` | Python deps broken | Reinstall missing packages |\n| `Connected to Telegram (polling mode)` | OK | — |\n| `set_my_commands OK` | OK | — |\n| `ImportError: cannot import name '\u003CX>' from '\u003Cmodule>'` | Version skew — daemon running pre-update code against post-update files | See Layer 1.5 |\n\n## Layer 2.4 — Version Skew: gateway process predating a code update\n\n**Symptom:** every inbound message fails with `ImportError: cannot import name\n'X' from 'Y'`, naming a symbol that plainly exists in the file on disk. Multiple\nunrelated symbols fail at once, from modules that import fine interactively.\n\n**Cause:** `hermes update` (or a git checkout of a new tag) replaces the source\ntree while the long-running gateway keeps the OLD modules already imported in\nmemory. The process holds a mix of old code and new files: old call sites\nimport symbols the new layout moved or renamed. Nothing on disk is broken.\n\n**Diagnose — compare process start time against the update, not against the file:**\n\n```bash\n# When did the running gateway start?\nps -o pid,lstart,etime -p \"$(pgrep -f 'hermes_cli.main gateway run' | head -1)\"\n\n# When was the source last replaced?\nstat -c '%y' ~\u002F.hermes\u002Fhermes-agent\u002Fhermes_cli\u002Fproviders.py\ngit -C ~\u002F.hermes\u002Fhermes-agent log -1 --format='%ci %s'\n```\n\nIf the process started BEFORE the update, version skew is the diagnosis.\n\n**Confirm the disk side is actually healthy** — this is the step that separates\nskew from a real breakage:\n\n```bash\n~\u002F.hermes\u002Fhermes-agent\u002Fvenv\u002Fbin\u002Fpython -c \"from hermes_cli.providers import LLAMACPP_ALIASES; print('imports OK')\"\n~\u002F.hermes\u002Fhermes-agent\u002Fvenv\u002Fbin\u002Fpython -c \"from tools.mcp_tool_common import mcp_server_enabled; print('imports OK')\"\n```\n\nIf those succeed, the fix is a restart, not a code change or reinstall:\n\n```bash\nsystemctl --user restart hermes-gateway.service\n```\n\n**Verify by scoping errors to the new process, never by counting the whole log.**\n`grep -c ImportError gateway.log` is cumulative and will show the historical\nfailures forever:\n\n```bash\nawk '\u002F\u003Crestart timestamp>\u002F,0' ~\u002F.hermes\u002Flogs\u002Fgateway.log | grep -c ImportError   # must be 0\ngrep -E \"telegram connected\" ~\u002F.hermes\u002Flogs\u002Fgateway.log | tail -1\n```\n\nHealthy reconnect looks like:\n```\n[Telegram] Telegram polling confirmed healthy: getUpdates progressing (generation 1)\ngateway.run: ✓ telegram connected\n```\n\n**PITFALL — an update procedure that does not restart the gateway leaves the\nsystem silently broken.** Any task that replaces Hermes source (update, tag\ncheckout, dependency sync) MUST end with the gateway restart plus the\npost-restart connectivity check. Report the update as incomplete until\n`✓ telegram connected` appears with a process start time newer than the update.\n`hermes doctor` does NOT catch this — it inspects on-disk config, not the\nrunning process's loaded modules.\n\n## Layer 2.5 — Model\u002FProvider Failure Diagnosis\n\nWhen a session shows `⚠️ The model provider failed after retries. I kept raw provider details out of chat; check gateway logs for diagnostics.` — the raw details ARE in the logs, the chat message just hides them by design:\n\n```bash\n# errors.log (NOT gateway.log) carries per-attempt provider details\ngrep -i \"api call failed\" ~\u002F.hermes\u002Flogs\u002Ferrors.log | tail -20\ngrep -i \"API call failed\" ~\u002F.hermes\u002Flogs\u002Ferrors.log | grep \"$(date +%F)\" | tail -20   # solo hoy\n```\n\nEach failed attempt logs `error_type=...` + `summary=...` for the exact provider\u002Fmodel. Taxonomy:\n\n| error_type | summary signal | Meaning | Action |\n|---|---|---|---|\n| `APIStatusError` | `HTTP 402 ... Insufficient Balance` | **Provider account balance exhausted (user-actionable)** | Top up the provider account \u002F swap API key |\n| `InternalServerError` | `HTTP 500: {\"type\":\"Router.Unavailable\",\"modelID\":...}` | Provider router can't serve that model right now (transient) | Retry \u002F wait \u002F temporarily switch model in config.yaml |\n| `InternalServerError` | `HTTP 500: Internal server error` | Provider-side outage | Retry with backoff (Hermes already retries 3x) |\n| `APIConnectionError` | `Connection error.` | Provider unreachable (network\u002Foutage) | Retry; `curl -sI \u003Cbase_url>`; if ALL providers fail it's network |\n\nWorkflow:\n1. Get the timestamp of the failed turn from `agent.log`: `grep \"model provider failed\" ~\u002F.hermes\u002Flogs\u002Fagent.log* | tail -5`\n2. Grep `errors.log` around that timestamp for `API call failed` → read `error_type` + `summary`\n3. Classify: **user-actionable** (402 balance, 401 bad key — the user must act) vs **transient** (500\u002Fconnection — retry works, no user action)\n4. Check whether the SAME error persists now (`tail errors.log`) — a one-off 500 vs a persistent `Router.Unavailable` are different answers\n5. Also watch `agent.credential_pool: no available entries (all exhausted or empty)` — credential pool empty compounds the failure\n\nReal case 2026-08-02\u002F03: chat error caused first by `HTTP 402 Insufficient Balance` (opencode.ai account empty — user needed to top up), then the next day by `HTTP 500 Router.Unavailable modelID=deepseek-v4-flash` (provider-side, transient). Two different causes on consecutive days, both visible ONLY in errors.log.\n\n## Layer 2.6 — Hermes Execution Health Audit (logs + entorno, verificado 2026-08-05)\n\nCuando el usuario pide \"revisa los logs y todo alrededor para mejorar la ejecución de Hermes\", la\nauditoría NO es solo del gateway — es de todo el proceso. Pasos que producen hallazgos accionables:\n\n```bash\n# 1. Tamaño y rotación de logs — agent.log rota a ~5MB; errors.log a ~2MB\nls -la ~\u002F.hermes\u002Flogs\u002F\n# 2. Tipos de error en agent.log (top tool failures del turno actual)\ngrep \"WARNING\" ~\u002F.hermes\u002Flogs\u002Fagent.log | grep -oE \"Tool [a-z_]+ returned error|Tool [a-z_]+: |agent\\.[a-z_]+:\" | sort | uniq -c | sort -rn | head -8\n# 3. Caídas del gateway — gateway-exit-diag.log registra el traceback del exit code\ntail -20 ~\u002F.hermes\u002Flogs\u002Fgateway-exit-diag.log\njournalctl --user -u hermes-gateway.service --since \"3 days ago\" | grep -cE \"Started|Stopped\"   # reinicios\n# 4. Recursos: RAM del gateway y pico histórico (pico 28GB en esta máquina = posible OOM previo)\nsystemctl --user status hermes-gateway.service | grep -E \"Active|Memory|Tasks\"\n# 5. state.db creciendo (sesiones\u002Fmensajes almacenados — 2.6GB\u002F245K msgs en esta máquina)\nls -lh ~\u002F.hermes\u002Fstate.db\n```\n\n**Hallazgo recurrente — background curator ruidoso (34+ errores\u002Fturno):** el self-improvement\nreview corre después de cada turno e intenta parchear skills vía `skill_manage`, pero se le NEGA\nsistemáticamente, llenando errors.log:\n```\nRefusing background curator patch for skill 'X': the skill records show it is not agent-created\n(created_by=None). Manually authored skills are off-limits... \u002F current SKILL.md content has not\nbeen loaded in this review turn...\n```\n- **Causa**: skills escritas a mano (created_by=None en `.usage.json`) o el review no carga el\n  SKILL.md antes de parchear → el guard las rechaza. No es un fallo del skill, es el review\n  intentando auto-mejorar contenido que no le corresponde.\n- **Impacto**: llamadas de API fallidas por turno + logs inundados. Si `errors.log` está lleno de\n  este patrón, es la fuente de ruido #1.\n- **Fix propuesto**: configurar `auxiliary.background_review` a un modelo barato explícito o\n  desactivar el auto-parchado de skills manuales (dejar solo captura de memoria).\n- **PITFALL de patch a skills manuales**: el mismo guard bloquea `skill_manage patch\u002Fedit` sobre\n  skills con `created_by=None` (ej. skills creadas antes de que el registro `.usage.json`\n  existiera, o cuyo create inicial falló y dejó el registro huérfano). No es un error de la\n  operación — es el guard anti-curation. Workaround: si el skill está protegido pero el\n  aprendizaje es importante, documentarlo en un skill editable (los que tienen\n  `author: Hermes Agent` en frontmatter) o crear uno nuevo.\n\n**gateway exit code 75 (verificado varias veces 2026-08-01..03):** `SystemExit(75)` en\n`gateway\u002Frun.py:22936` — el gateway se cae con este código cuando el runner aborta. Diagnóstico\nen `gateway-exit-diag.log` (traceback completo con PID + timestamp). Los reinicios frecuentes\n(7 en 3 días) con pico de RAM de 28GB apuntan a OOM por delegaciones masivas o compresión de\ncontexto — vigilar memoria antes de asumir problema de código.\n\n**state.db 2.6GB (245K mensajes, 3.4K sesiones):** cada `session_search` indexa todo el DB;\nbackups lentos. Mantenimiento: podar sesiones >60 días (exportar a Xavier antes) + `VACUUM`.\n\n## Layer 1.5 — Version Skew After an In-Place Update\n\nThe highest-yield check when a gateway breaks right after `hermes update`: the\nrunning process holds the OLD code in memory while the files on disk are NEW.\nEvery inbound message then dies on an import that no longer resolves.\n\nSignature in `~\u002F.hermes\u002Flogs\u002Fgateway.log`:\n\n```\nImportError: cannot import name 'LLAMACPP_ALIASES' from 'hermes_cli.providers'\nImportError: cannot import name 'mcp_server_enabled' from 'tools.mcp_tool_common'\n```\n\nRead as a module-move in the new release, not as a broken install.\n\n**Triage — compare the process start time against the file mtime:**\n\n```bash\nps -o pid,lstart,etime -p $(pgrep -f \"hermes_cli.main gateway run\" | head -1)\nstat -c '%y' ~\u002F.hermes\u002Fhermes-agent\u002Fhermes_cli\u002Fproviders.py\n```\n\nProcess started hours BEFORE the update ⇒ skew. Confirm the symbol is fine with\ncurrent code before touching anything:\n\n```bash\n~\u002F.hermes\u002Fhermes-agent\u002Fvenv\u002Fbin\u002Fpython -c \\\n  \"from hermes_cli.providers import LLAMACPP_ALIASES; print('imports OK')\"\n```\n\nIf that succeeds, the fix is a restart and nothing else:\n\n```bash\nsystemctl --user restart hermes-gateway.service\n```\n\nThen verify the NEW process, and prove the skew is gone:\n\n```bash\nps -o lstart= -p $(pgrep -f \"hermes_cli.main gateway run\" | head -1)   # fresh\ngrep -a \"✓ telegram connected\" ~\u002F.hermes\u002Flogs\u002Fgateway.log | tail -1\n# count errors only AFTER the new start time — a whole-file count stays non-zero forever\nawk '\u002F\u003Cnew start timestamp>\u002F,0' ~\u002F.hermes\u002Flogs\u002Fgateway.log | grep -c ImportError   # expect 0\n```\n\n**PITFALL:** `grep -c ImportError gateway.log` is cumulative and never returns 0\nafter an update. Always scope the count to the region after the restart, or you\nwill report a resolved fault as a live one.\n\n**PITFALL:** `gateway.log` may be detected as a binary file, making `grep` print\n`coincidencia en fichero binario` and silently match nothing. Add `-a` (as\nabove) whenever you grep a log that has grown large or contains non-UTF8 bytes.\n\n**Prevent it:** stop the gateway before updating, or restart it immediately\nafter, as one atomic task. An update that leaves a long-lived daemon running\nagainst replaced source will break every platform adapter until that restart.\n\n## Layer 3 — Systemd Service Config\n\n```bash\n# View the full service unit\nsystemctl --user cat hermes-gateway.service\n\n# Check for these critical fields:\n# - WorkingDirectory: must point to CURRENT python version's site-packages\n# - Environment=PATH: must reference correct python version\n# - ExecStart: correct python binary path\n```\n\n### Common Service File Issues\n\n**WorkingDirectory pointing to old Python version:**\n```\n# WRONG (stale python3.13):\nWorkingDirectory=\u002Fhome\u002Fuser\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Flib\u002Fpython3.13\u002Fsite-packages\n\n# RIGHT (current python3.14):\nWorkingDirectory=\u002Fhome\u002Fuser\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Flib\u002Fpython3.14\u002Fsite-packages\n```\n\n### Fix\n```bash\n# Edit the service file\nnano ~\u002F.config\u002Fsystemd\u002Fuser\u002Fhermes-gateway.service\n# Then:\nsystemctl --user daemon-reload\nsystemctl --user restart hermes-gateway.service\n```\n\n## Layer 4 — Python Environment (uv Tool)\n\n### Detect Dual-Version Site-Packages\n\nWhen uv upgrades the hermetic Python inside a tool, old version-specific site-packages can linger and cause import shadowing (Python resolves modules from the stale directory first).\n\n```bash\n# Check which python versions exist in the uv tool\nls ~\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Flib\u002F\n\n# Check actual python version of the tool's binary\n~\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Fbin\u002Fpython3 --version\n\n# If you see python3.13\u002F AND python3.14\u002F, the old one is stale\n# Fix: ensure systemd WorkingDirectory points to the CURRENT version\n# Fix: install missing packages into the current version's site-packages\n```\n\n**Root cause:** Python adds the current working directory to `sys.path` as `''` (empty string). If `WorkingDirectory` points to an old pythonX.Y\u002Fsite-packages, Python finds packages there first — even when `sys.path` doesn't explicitly include that directory.\n\n### Install Missing Packages\n\nWhen the correct site-packages directory lacks packages that only exist in a stale version:\n\n```bash\nuv pip install --python ~\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Fbin\u002Fpython3 \\\n  pydantic pydantic-core \\\n  python-telegram-bot \\\n  anthropic openai \\\n  rich tenacity httpx[socks] edge-tts \\\n  certifi python-dotenv fire pyyaml \\\n  jinja2 psutil tabulate termcolor wcwidth\n```\n\n### Verify\n```bash\n~\u002F.local\u002Fshare\u002Fuv\u002Ftools\u002Fhermes-agent\u002Fbin\u002Fpython3 -c \"\nimport pydantic_core\nimport pydantic\nimport telegram\nimport openai\nprint('pydantic_core:', pydantic_core.__version__)\nprint('pydantic:', pydantic.__version__)\nprint('ALL OK')\n\"\n```\n\n## Layer 5 — Plugin\u002FAdapter Config\n\n### Telegram `allow_from` Format\n\nHermes 0.19.0+ uses a plugin-based Telegram adapter (`plugins\u002Fplatforms\u002Ftelegram\u002F`). The `allow_from` key is read from `config.yaml`'s `telegram:` section and bridged into the plugin's `extra` dict via `gateway\u002Fconfig.py` (~line 1472-1473 → `extra.update(bridged)` at ~line 1524).\n\n**Critical:** The adapter iterates over `allow_from` to build an allowed set. Four formats can arrive:\n\n| YAML config | Python type | Iteration behavior |\n|---|---|---|\n| `allow_from: [2076598024]` | `list[int]` | ✅ One iteration per ID |\n| `allow_from:\\n  - '2076598024'` | `list[str]` | ✅ One iteration per ID |\n| `allow_from: 2076598024` | `int` | ❌ `TypeError: 'int' object is not iterable` (traceback at `adapter.py:1019`) |\n| `allow_from: '2076598024'` | `str` | ❌ Iterates character-by-character (`'2'`, `'0'`, ...) |\n| `allow_from: '[\"2076598024\"]'` | `str` | ❌ Produces literal set `{'[\"2076598024\"]'}` — never matches |\n\nThe last row is the **`hermes config set` pitfall**: `hermes config set telegram.allow_from '[id]'` stores the value as a **literal quoted string** `'[\"id\"]'` in YAML, **NOT** as a native YAML list. Always edit `config.yaml` directly or use `sed`\u002FPython to write the YAML list format properly.\n\n**Preferred config format:**\n```yaml\ntelegram:\n  allow_from:\n    - '2076598024'       # YAML list with one quoted string per user ID\n```\n\nMultiple users:\n```yaml\ntelegram:\n  allow_from:\n    - '2076598024'\n    - '1234567890'\n```\n\n**Adapter patch** (shown with line numbers from `adapter.py:1016-1020`):\n\nThe original code assumes `allow_from` is always a list:\n```python\n# Original (broken for int\u002Fstr):\nadapter_allow_from = self.config.extra.get(\"allow_from\")\nif adapter_allow_from is not None:\n    allowed = {str(u).strip() for u in adapter_allow_from if str(u).strip()}\n    return user_id in allowed or \"*\" in allowed\n```\n\nDefensive fix — wrap non-list values before iterating:\n```python\nadapter_allow_from = self.config.extra.get(\"allow_from\")\nif adapter_allow_from is not None:\n    if isinstance(adapter_allow_from, str):\n        adapter_allow_from = [adapter_allow_from]       # str → list\n    elif not isinstance(adapter_allow_from, (list, tuple)):\n        adapter_allow_from = [str(adapter_allow_from)]  # int\u002Ffloat → list\n    allowed = {str(u).strip() for u in adapter_allow_from if str(u).strip()}\n    return str(user_id) in allowed or \"*\" in allowed\n```\n\n**Key details:**\n- Use `str(user_id)` for comparison — Telegram may pass `user_id` as int while the allowed set contains strings\n- After patching `adapter.py`, **restart the gateway** via `systemctl --user restart hermes-gateway.service` — the code change is hot only after restart\n- The config → plugin bridging happens in `gateway\u002Fconfig.py` during gateway startup; changes to `config.yaml` only take effect after restart\n- If you still get errors after fixing config, check that `systemctl --user daemon-reload` ran if you edited the service file\n\n### Config → Plugin Bridging\n\nThe gateway config loader (`gateway\u002Fconfig.py`) reads top-level platform blocks from `config.yaml` and bridges specific keys into the plugin's `extra` dict. Bridged keys include:\n\n`allow_from`, `allowed_chats`, `group_allowed_chats`, `allowed_topics`, `require_mention`, `dm_policy`, `group_policy`, `gateway_restart_notification`, `typing_indicator`, `channel_prompts`, `free_response_channels`, `mention_patterns`\n\n## Full Debugging Sequence\n\nWhen Telegram (or any platform) stops responding:\n\n```\n1. systemctl --user status hermes-gateway.service\n   → Is it running? If not, start it.\n\n2. tail -50 ~\u002F.hermes\u002Flogs\u002Fgateway.log | grep -E \"ERROR|Traceback|Blocked|unauthorized\"\n   → What errors exist?\n\n3. If `TypeError: 'int' object is not iterable` at `adapter.py:1019`:\n   → `allow_from` in config.yaml is an unquoted integer parsed by YAML as `int`\n   → Fix: change to proper YAML list format\n   → **WARNING:** `hermes config set telegram.allow_from '[id]'` stores a JSON-stringified\n     literal string `'[\"id\"]'`, NOT a YAML list. Edit config.yaml directly.\n   → If the adapter doesn't handle all formats (str\u002Fint\u002Flist), patch adapter.py\n   → **Always restart gateway after adapter patch** (`systemctl --user restart hermes-gateway.service`)\n\n4. If \"Blocked unauthorized user YOUR_ID\":\n   → Check telegram.allow_from format in config.yaml (same fix as above)\n   → Ensure the user_id in config matches the Telegram user ID exactly\n\n5. If pydantic_core ModuleNotFoundError:\n   → Check uv tool python version vs site-packages ABI\n   → Install missing packages for current python version\n\n6. If code loads from wrong site-packages:\n   → Check systemd service WorkingDirectory\n   → Update to point to current python version\n\n7. Restart gateway and test.\n```\n\n## Verification\n\n```bash\n# Send a test message via Telegram and check logs\ngrep -E \"inbound message|Processed message|Agent error\" ~\u002F.hermes\u002Flogs\u002Fgateway.log | tail -5\n```\n",{"name":5,"author":24,"version":25,"platforms":26,"description":7},"Hermes Agent","1.0.0",[27],"linux","main","688810f2886d212632acf81e532b24e483dd8b98","skills\u002Fdevops\u002Fhermes-gateway-troubleshooting","https:\u002F\u002Fgithub.com\u002Fiberi22\u002Fskills","https:\u002F\u002Fgithub.com\u002Fiberi22","Senior WEB Full-Stack","2026-10-12 02:46:21",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add iberi22\u002Fskills --skill hermes-gateway-troubleshooting",{"tool":41,"command":42},"bunx","bunx skills add iberi22\u002Fskills --skill hermes-gateway-troubleshooting",{"tool":44,"command":45},"pnpm","pnpm dlx skills add iberi22\u002Fskills --skill hermes-gateway-troubleshooting","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fiberi22\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fdevops\u002Fhermes-gateway-troubleshooting\u002FSKILL.md",true,[],[50,56,63,69,75,81],{"skillNo":51,"name":52,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":53,"categories":54,"keywords":55,"publishedAt":19},"01M72TWQEQD46ZC9TRBAABF78P","web-design-guidelines","https:\u002F\u002Fgithub.com\u002Fiberi22\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fweb\u002Fweb-design-guidelines\u002FSKILL.md",[],[],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":19},"01M72TWMQFTYD08DGXVAXRPPEB","blocked-page-recovery","Use when a fetch fails: 403\u002F429, paywall, WAF, bot wall. Recovers the page via archives and provenance-checked fallbacks.","https:\u002F\u002Fgithub.com\u002Fiberi22\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fweb\u002Fblocked-page-recovery\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":66,"categories":67,"keywords":68,"publishedAt":19},"01M72TVQMS3HCTX18C8T97WTQQ","studio-coach","https:\u002F\u002Fgithub.com\u002Fiberi22\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Ftools\u002Fstudio-coach\u002FSKILL.md",[],[],{"skillNo":70,"name":71,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M72TVNA2PR9RZW0ZAXYP1G6J","src-generator","https:\u002F\u002Fgithub.com\u002Fiberi22\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Ftools\u002Fsrc-generator\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":19},"01M72TVEJHHB6SAT3R3TY9MJKS","sqlite-pro","https:\u002F\u002Fgithub.com\u002Fiberi22\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Ftools\u002Fsqlite-pro\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M72TVC2EVBHQHCX96VTTYKB8","rigor-evaluator","Use when you need a quick heuristic rigor score (10 criteria, 0-2 each) for the features in a .gitcore\u002Ffeatures.json file, to decide which features are ready to ship.","https:\u002F\u002Fgithub.com\u002Fiberi22\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Ftools\u002Frigor-evaluator\u002FSKILL.md",[],[]]