Add OpenCode Go usage sample

[4xvgal]

Context of Contribution

• Others

Summary of the Proposal

Add an OpenCode Go usage sample under docs/samples/opencode-go/ that
tracks rolling, weekly, and monthly OpenCode Go subscription usage in
RunCat Neo's Custom Metrics card and Metrics Bar.

Reason for the new feature

OpenCode Go subscribers ($10/month) cannot see their usage limits from
the terminal — they must visit the web dashboard. This sample follows
the same launchd + JSON snapshot pattern as the existing Bitcoin sample,
so users can monitor their Go usage directly in their macOS menu bar
alongside system metrics.

Checklist

• I have read the CONTRIBUTING.md and agree to follow it.
• This PR does not contain commits of multiple contexts.
• Code follows proper indentation and naming conventions.
• Implemented using only APIs that can be submitted to the App Store.

[Kyome22]

@4xvgal
Could you please attach a screenshot showing how it actually looks on your screen?

[4xvgal]

It looks like this! design came from claude example

[Kyome22]

Thanks for the contribution! Before getting into the code, I want to raise a concern about the overall direction, because I think that needs to be resolved before this can be considered for merging regardless of code quality.

The existing samples under docs/samples/ all rely on either a documented public API (Bitcoin → CoinGecko) or an official lifecycle hook exposed by the host tool (Claude Code statusLine, Codex Stop hook). This sample departs from that pattern in a way I'm not comfortable shipping as an "official" example:

• Session cookie extraction. Instructing users to open DevTools, copy opencode.ai's auth cookie, and drop it into a dotfile is a security posture I don't want to normalise in this repo. That cookie is bearer-equivalent — if it leaks, it's effectively account takeover on opencode.ai. chmod 600 doesn't change that.
• Scraping a private dashboard. opencode.ai/workspace/:id/go isn't a documented API. It will break the moment the page markup changes, and it sits in a grey area with respect to OpenCode's ToS. I'd rather RunCat Neo not nudge users down that path in its own docs.
• Third-party CLI as a hard dependency. The sample requires cloning + go build-ing a small single-file scraper from a personal sourcehut repo (default branch is batman, not main), which additionally has a hardcoded fallback workspace ID baked into the binary. That is a lot of fragility to layer a sample on top of.

OpenCode is open source and, similar to Codex, keeps a local session store on disk. The far more natural sample here would mirror docs/samples/codex/ — a lifecycle hook (or a launchd-scheduled script) that reads OpenCode's local session data, with no cookie, no scraper, and no third-party binary. That would slot in cleanly next to the three existing samples.

Would you be open to redirecting the PR along those lines? As it stands I don't think the scrape-based approach is one we can adopt as an official sample. Happy to discuss if you see it differently.

[Kyome22]

Separately from the direction question above, I also went through the current code and found a few quality issues worth flagging, in case the sample lands somewhere similar.

Bugs

• plist Label / bootout command mismatch. The plist's <Label> is dev.runcat.opencode-go-sample, but README step 6 copies the file to ~/Library/LaunchAgents/dev.runcat.opencode-go.plist (renaming it) and the Stopping section runs launchctl bootout gui/\$(id -u)/dev.runcat.opencode-go. bootout takes the Label, not the filename, so the stop command as written will not find the agent. The Bitcoin sample keeps file / Label / bootout all consistent as dev.runcat.bitcoin-sample; the same needs to hold here.
• Missing OPENCODE_WORKSPACE silently pulls someone else's data. If the env var is unset the script just omits -workspace, and the upstream CLI falls back to a workspace ID hardcoded into its binary (wrk_01KT241XYZ…, per the README's own Troubleshooting section). The script should fail fast when \$workspace is empty rather than render a stranger's usage in the user's menu bar.
• null values from jq degrade silently. jq -r '.rolling.percent' on a missing field emits the literal string null, awk coerces it to 0, and the card cheerfully shows 0% with normalizedValue: 0.0000 as if it were real data. Consider jq -e (or an explicit null check) so the script errors out rather than lying.
• lastUpdatedDate isn't validated. It's copied straight from the upstream .fetched_at — if the upstream ever emits a non-ISO-8601 string, RunCat can't parse it. A date -u +%Y-%m-%dT%H:%M:%SZ fallback keeps the card honest.

Unused code

• plan=\$(echo \"\$usage\" | jq -r '.plan') on line 39 is assigned and never referenced. This repo's convention (see CLAUDE.md) is to remove variables/functions that end up unused.

Nits

• The five separate echo \"\$usage\" | jq -r … invocations can be consolidated into a single jq call that emits all fields at once — cheaper and easier to read.
• args=\"\$args -workspace \$workspace\" and unquoted \$args expansion depend on word-splitting; positional parameters (set -- ... -workspace \"\$workspace\") would be safer against paths with spaces. Not an actual hazard with the current defaults, just idiomatically cleaner.
• Given how fragile the data source is, the README would benefit from a warning similar to the one at the top of docs/samples/codex/README.md ("format may change; adjust the parser if it breaks").

None of these are large to fix, but flagging so they don't carry over if the sample is redirected to a local-session-based approach.