소스 정보
- 저장소
- lukemcqueen/hermes-cortex
- 최근 소스 활동
- 2026년 8월 25일 00:37
- 감지된 SKILL.md 언어
- 영어
- 스타
- 4
- 포크
- 1
설치 방법
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
소스 파일 검토
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
메뉴
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/lukemcqueen/hermes-cortex --skill hermes-model-config명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SKILL.md 표시 중
| name | hermes-model-config |
| description | Use when sessions run the wrong model or config regresses. |
| version | 1.0.0 |
| category | devops |
| author | Titus |
| license | MIT |
| platforms | ["linux","macos"] |
Trigger: "sessions use the wrong model", "model is X but should be Y",
deepseek-chat/deepseek-flash questions, ⚠ Deprecated .env settings detected
warning, cron model overrides, model.default regressions after hermes update.
deepseek-v4-flash — set in config.yaml
model.default (provider deepseek, base_url https://api.deepseek.com/v1).
Also encoded as DEFAULT_MODEL=deepseek-v4-flash in ~/.hermes-cortex/.env.deepseek-chat is LLM-cron-ONLY. It must never be the session default.
Crons pin it per-job: "model": "deepseek-chat" in ~/.hermes/cron/jobs.json
(see LLM_CRON_MODEL=deepseek-chat in ops/scripts/install-crons.sh and
ops/install/cron-manifest.yaml). A config.yaml default of deepseek-chat is
the regression symptom.hermes config get model.default (+ model.provider, model.base_url).
The session banner snapshots the model at session START — a mid-session
hermes update changes config underneath it (observed 2026-08-18: banner
showed deepseek-v4-flash while config already said deepseek-chat).stat -f '%Sm %N' ~/.hermes/config.yaml and tail ~/.hermes/logs/update.log
— hermes update rewrites config.yaml AND ~/.hermes/.env in the same
second. To rule out a migration, grep the key in
~/.hermes/hermes-agent/hermes_cli/config_migrations.py (model.default had
NO migration — it was a stale manual setting).grep -n '"model"' ~/.hermes/cron/jobs.json — per-job
model field; model_snapshot records what it actually ran as. no_agent
jobs (script-based) ignore model entirely.grep -rn '<model-id>' ~/hermes-cortex --include='*.sh' --include='*.yaml' --include='*.py' —
install scripts/manifests show the intended per-role model split.hermes config set model.default <model-id> — NEVER hand-edit
config.yaml (stray indent breaks the live gateway; hermes-agent hard
invariant). provider / base_url are untouched by this command._resolve_gateway_model (gateway/run.py) reads
model.default fresh per session from raw YAML.hermes chat -q "Reply with exactly: OK" — note the
Session: id in output.sqlite3 ~/.hermes/state.db "SELECT id, model, billing_provider, billing_base_url FROM sessions WHERE id='<session-id>';"
⚠️ sessions table has model + billing_provider — there is NO provider
column; billing_provider is the provider.warn_deprecated_cwd_env_vars
(hermes_cli/config.py) and checks process env for MESSAGING_CWD /
TERMINAL_CWD — it can false-positive on Hermes's own launch-dir bridge
(the message says "found in .env" even when the var came from Hermes).hermes doctor, which scans the .env file
(hermes_cli/doctor.py _DEPRECATED_ENV_VARS): TERMINAL_CWD →
terminal.cwd, MESSAGING_CWD → terminal.cwd,
HERMES_TOOL_PROGRESS / HERMES_TOOL_PROGRESS_MODE → display.tool_progress,
QQ_HOME_CHANNEL / QQ_HOME_CHANNEL_NAME → QQBOT_HOME_CHANNEL[_NAME].hermes update auto-comments deprecated .env entries. Check for ACTIVE
leftovers with:
grep -nE '^(TERMINAL_CWD|MESSAGING_CWD|HERMES_TOOL_PROGRESS|HERMES_TOOL_PROGRESS_MODE|QQ_HOME_CHANNEL|QQ_HOME_CHANNEL_NAME)=' ~/.hermes/.envterminal.cwd is PER-PROJECT: keep terminal.cwd: .
(launch dir) on multi-project hosts. Do NOT set a global project path just
to silence the warning.state.db sessions table: billing_provider, not provider (query error).
The .env deprecated entry may already be commented by a recent hermes update — verify before "removing" anything; an empty grep result means the
warning is already neutralized.
model ids evolve: verify the exact id against
grep -n '"model"' ~/.hermes/cron/jobs.json and ~/.hermes-cortex/.env
before assuming a shorthand ("deepseek-flash" = deepseek-v4-flash).
2026-08-18 deepseek-chat regression: investigation trail + verification output.