Two client-side fixes against the covert-gpt LiteLLM gateway, both traced
from opencode throwing intermittent ContextWindowExceededError 400s.
opencode (mcp-config/sync.sh):
spec_for() mapped LiteLLM's /model/info onto opencode's schema 1:1, but the
two mean different things by "output". LiteLLM's max_output_tokens is the
whole window (vLLM caps TOTAL context, not output separately, which is
correct for LiteLLM); opencode's limit.output is a RESERVE it subtracts from
context to pick a compaction point. Copying one to the other told opencode to
reserve the entire 262144 window, so it targeted an input of context-output
and the server counted one token more than opencode did -- landing on
context+1 and a hard 400, every time (measured: 196609 in + 65536 out =
262145 vs a 262144 window; the retries shrinking max_tokens by 65 never
helped because the input grew to compensate).
Capping the reserve alone would NOT have fixed it -- opencode would still sit
exactly on the boundary, just at a different input size. So: cap the reserve
(OUTPUT_RESERVE_CAP) and hold a margin back from context
(CONTEXT_SAFETY_MARGIN). Generated at sync time, so it fixes every gateway
model at once, including third-party ones.
hermes (hermes/gateway.yaml, hermes/install.sh):
Hermes reached the gateway by pretending it WAS OpenRouter -- the LiteLLM key
in OPENROUTER_API_KEY, the gateway in OPENROUTER_BASE_URL. Requests routed
fine, but fetch_openrouter_models hardcodes openrouter.ai/api/v1/models and
ignores OPENROUTER_BASE_URL, so the picker listed OpenRouter's ~342-model
catalog -- models this gateway cannot serve -- while the gateway's own models,
including cg-1's, never appeared at all.
Replaced with a named custom provider. install.sh now merges gateway.yaml into
~/.hermes/config.yaml, so a fresh box gets a Hermes that can actually talk to
the gateway instead of one pointed at nothing. The merge is idempotent and
narrow: only the covert-gpt entry is touched, model.default/provider are
seeded only when unset, and a box still on the masquerade is migrated off it
(fingerprint: provider openrouter + base_url pointing at this gateway; a
genuine OpenRouter user never has that, and is left alone).
No secrets: gateway.yaml names the key via key_env, and mcp-config/sync.sh is
what exports it. Hermes needs no context safety margin -- it omits max_tokens
entirely, so vLLM bounds output by whatever context is left.
Also folds in the in-flight hermes bootstrap wiring.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Models and MCP servers are discovered from the LiteLLM gateway at sync time
(/v1/models, and the aggregate /mcp/ endpoint). Skills have no gateway side to
query: LiteLLM's /v1/skills is a passthrough to Anthropic's Skills API, which
lists skills on an Anthropic account rather than serving skill files, and
/public/skill_hub is a marketplace catalog. So the repo is the source of truth
and the tree is fetched whole, from the repo tarball.
Deploy to ~/.config/mcp-sync/skills rather than ~/.config/opencode/skills so
the atomic replace can never delete a skill the user wrote by hand; opencode
picks the tree up via skills.paths, which expands {env:HOME} — verified — so
the shared config stays identical for every user.
Same throttle and non-fatal discipline as the config sync: a fetch that yields
no skills/ leaves the last-good tree in place instead of emptying it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
~/.zshrc is managed and overwritten on every bootstrap, so there was
nowhere to put a machine-local key. The managed rc now sources
~/.zshrc.local last (so it can override anything) and bootstrap never
touches that file. zshrc.local.example documents the variables modules
look for: GITEA_KEY, LITELLM_KEY, OP_SERVICE_ACCOUNT_TOKEN.
Secrets stay out of the repo — zsh/zshrc is shared config applied to
every box.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
kind never worked reliably against rootless Docker on Kasm, even with the
cgroup delegation from docker-rootless-cgroup/. So install full Docker
where we can, and flip bin/'s kind gate from "rootless is available" to
"a rootful daemon answers" — kind is simply skipped on Kasm now.
docker/install.sh no-ops when a rootless daemon is already running so it
never fights Kasm's setup. It runs before bin/ in bootstrap so the kind
gate can see the daemon it just started. zsh/zshrc re-execs via newgrp
when the account gained the docker group but the live shell predates it.
docker-rootless-cgroup/ keeps its delegation (still useful for
per-container resource limits); only its kind-centric comments change.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
New bin/install.sh downloads self-contained release binaries (ripgrep, fd,
bat, fzf, jq) into ~/.local/bin — no sudo, nothing outside $HOME, so they
persist across Kasm redeploys with no reinstall. Statically linked where
available, so they keep working even after the root fs is wiped. Idempotent:
pinned versions already present are skipped.
lib/fetch.sh adds arch detection + tarball/raw binary download helpers.
zsh/zshrc puts ~/.local/bin on PATH; bootstrap.sh runs the module.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Cross-distro (RHEL/Debian) VM setup: zsh + oh-my-zsh with a self-healing
bash fallback for environments that reset the root filesystem or hardcode
bash for new terminals (e.g. Kasm Workspaces), plus git identity and
HTTPS credential-store configuration.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
noconnorandClaude Opus 4.8