Repository navigation
Releases
v0.2.3 — OpenCode 2 benchmark results
Updated README with measured OpenCode 2.0.26 results for GPT 6 Sol and GPT 6.1 Sol (reasoning: low), tested on 2026-10-11 with Context Goblin 0.2.2. Observed total event token reductions: 33% and 48%, respectively. Historical OpenCode 1 results remain dated and separate. Added a resumable V2 benchmark runner and accounting tests; 98 tests pass. These are single-run measurements, not guaranteed repeatable or billing savings. No runtime plugin changes.
Context Goblin 0.2.2 — README cleanup
Reorganizes the README around installation, tool behavior, cache rules, optional features, and measured results. Removes repeated explanations and moves local development and benchmark commands into a dedicated guide. Plugin behavior is unchanged.
Context Goblin 0.2.1 — Documentation update
Clarifies the product description across GitHub and npm, removes repeated introductory text, and updates the installation example. Plugin behavior is unchanged.
Context Goblin 0.2.0 — OpenCode 2 support
Context Goblin 0.2.0 brings repository context caching to OpenCode 2, with continued support for OpenCode 1. Agents can reuse a compact project summary and code map before exploring the files needed for a task.
What's new
- OpenCode 2 integration: all six Context Goblin tools, cache and usage commands, and local token usage tracking now use the V2 plugin API.
- Optional terminal UI stats: view cache freshness, size, and coverage in a toast. Stats follow the selected session's project, including when it differs from the UI's startup directory.
- Safer TUI error handling: an unavailable session or cache error produces a message instead of showing another project's stats or failing the command.
- Updated setup instructions: the README covers V2 server and TUI configuration, local development, and avoiding duplicate plugin registrations.
Use with OpenCode 2
Add this entry to opencode.json or opencode.jsonc, then restart OpenCode:
{
"plugins": ["context-goblin@0.2.0"]
}For the optional TUI stats command, add the same entry to your OpenCode cli.json configuration. See the README for local setup and options.
Verification
Verified with OpenCode 2.0.26: the packaged plugin loads, both server commands register, and all six tools execute. All 90 tests and the release checks passed.
V2 model sessions and interactive TUI behavior still need validation. The existing A/B token savings measurements were collected on OpenCode 1.18.20; V2 performance has not been measured. Tool output compaction remains opt-in, and usage tracking reports OpenCode events rather than provider billing totals.
v0.1.23
Documentation and benchmark evidence update. No plugin runtime or protocol changes.
- Updated README and A/B reports with the completed 2026-10-08 comparison on OpenCode 1.18.20 and Context Goblin 0.1.22.
- Both models passed compatibility and efficiency checks: total token reductions were 27% for gpt-5.5 and 31% for gpt-5.6-sol in this run.
- Preserved the previous run, including the gpt-5.5 input-token regression, for transparent comparison.
- Clarified that repeated stability testing remains deferred; one-shot savings are not a repeatability guarantee.
Validation: 72 tests passed, typecheck, build, OpenCode integration smoke and report checks passed. No new paid model benchmark was run for this documentation-only release.
v0.1.22
Context Goblin 0.1.22 fixes unsafe source-link reads and stale project caches, and makes benchmark reporting stricter.
- Resolve source and metadata paths before reading; exclude links outside the workspace and links into denied paths.
- Detect source additions, edits, and deletions without Git, including allowed source-link targets. Track ignored source files in Git projects as well.
- Bound fallback/ignored source scans to 2,000 files and 32 MiB; treat unverifiable freshness as stale. Regenerate caches from older formats.
- Cache package script names and the presence of AGENTS.md instead of script bodies, instruction contents, or test descriptions. OpenCode remains responsible for loading native project instructions.
- Keep context_goblin_get as the recommended single-call flow; remove the conflicting four-call recommendation from the README.
- Run integration checks with the pinned OpenCode CLI. Add timeouts, stop model runs after provider errors, preserve completed reports, and reuse only matching-version benchmark data.
- Require every requested model to pass compatibility and all efficiency checks. Count only distinct files read through the read tool, and explain that the 6/6 answer score measures checklist coverage.
Validation: 72 local tests passed, typecheck/build/report checks passed, OpenCode 1.18.20 integration smoke passed, and the npm package dry run passed.
Fresh one-shot A/B on OpenCode 1.18.20:
| Model | Files saved | Uncached input saved | Total event tokens saved | Compatibility | Efficiency |
|---|---|---|---|---|---|
| gpt-5.5 | 53% | -28% | 7% | pass | fail |
| gpt-5.6-sol | 47% | 51% | 34% | pass | pass |
The gpt-5.5 input increase is reported as a failure. These single-run results do not establish repeatable savings across both models; the repeated stability benchmark remains deferred. Detailed answers and accounting are included in the repository reports.
v0.1.21
What's changed
- Added
context_goblin_get, a low-overhead single-call flow that:- checks cache freshness;
- refreshes missing or stale caches automatically;
- returns the safe project context in one tool call.
- Kept
context_goblin_status,context_goblin_refresh,context_goblin_read, andcontext_goblin_statsfor diagnostics and backward compatibility. - Updated OpenCode A/B benchmarks to use the new flow.
- Added strict efficiency reporting: any negative metric is reported as
fail, never as a successful result. - Updated README and benchmark reports with current OpenCode
1.18.20results. - Added tests for cache creation and stale-cache refresh behavior.
Verified results
57/57tests passing.- Typecheck and build passing.
- OpenCode live smoke passing.
gpt-5.5: 44% fewer file reads, 28% fewer input tokens, 47% fewer total event tokens.gpt-5.6-sol: 35% fewer file reads, 45% fewer input tokens, 35% fewer total event tokens.- Quality:
6/6for both models. - No secret leakage detected.
This is a small patch release with no breaking API changes.
v0.1.18
Context Goblin v0.1.17
Highlights
- Correct total token usage when provider records omit aggregate totals.
- Exclude future-dated events from usage statistics ranges.
- Restore compatibility with the legacy OpenCode TUI command fallback.
- Harden dependency and CI configuration.
- Update GitHub Actions to Node 24-compatible releases.
Full changelog: v0.1.16...v0.1.17