githits 0.11.1 → 0.11.3

package/.claude-plugin/marketplace.json

@@ -6,12 +6,12 @@
   },
   "metadata": {
     "description": "The code context layer for AI coding agents",
-    "version": "0.11.1"
+    "version": "0.11.3"
   },
   "plugins": [
     {
       "name": "githits",
-      "version": "0.11.1",
+      "version": "0.11.3",
       "description": "The code context layer for AI coding agents",
       "author": {
         "name": "GitHits"

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "githits",
-  "version": "0.11.1",
+  "version": "0.11.3",
   "description": "The code context layer for AI coding agents",
   "author": {
     "name": "GitHits"

package/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "githits",
-  "version": "0.11.1",
+  "version": "0.11.3",
   "description": "The code context layer for AI coding agents",
   "author": {
     "name": "GitHits"

package/.cursor-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "githits",
-  "version": "0.11.1",
+  "version": "0.11.3",
   "description": "The code context layer for AI coding agents",
   "author": {
     "name": "GitHits"

package/.plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "githits",
-  "version": "0.11.1",
+  "version": "0.11.3",
   "description": "The code context layer for AI coding agents",
   "author": {
     "name": "GitHits"

package/AGENTS.md

@@ -32,6 +32,18 @@ Philosophy: "Create architecture that is performant and easy to test"
 
 See `docs/guidelines/ARCHITECTURAL_GUIDELINES.md` for detailed planning checklist and design principles.
 
+## Tool output UX
+
+Human-readable tool text is an optimized product surface, not raw field
+serialization. Before editing it, inspect real output and preserve existing
+strengths: lead with the outcome, group related evidence, remove repetition and
+scaffolding, and retain stable follow-up locators, actions, and trust facts.
+Wrap free prose to the caller's width, keep formatter-authored punctuation ASCII
+while preserving backend Unicode, and never make color carry meaning. When CLI
+and MCP need the same information, share one formatter per tool with color and
+width as inputs; keep JSON lossless for machines. A complete data dump is not
+good output merely because it is complete.
+
 ## Testing
 
 Philosophy: "If it is not tested, it is likely broken"
@@ -40,7 +52,7 @@ Philosophy: "If it is not tested, it is likely broken"
 
 - Use `bun test` for running tests
 - Use `bun run smoke:mcp` and `bun run smoke:cli` when changing MCP tools, CLI commands, shared formatters, auth/error envelopes, or MCP/CLI parity behavior. These are live-capable local suites, not the normal unit suite; they must pass unauthenticated by validating auth handling, and provide deeper coverage when authenticated. After building, also run `bun run smoke:cli:built` and `bun run smoke:mcp:built` when changing smoke launch behavior or CI product validation; these secret-free modes execute `dist/cli.js` under Node.
-- Use `bun run agent:e2e` when changing MCP instructions, tool descriptions, or agent-facing tool behavior. This is a human/agent-driven qualitative eval, not a deterministic CI gate. Pick targeted workloads from `eval/agentic/README.md`; run both Claude and Codex for broad instruction changes when practical. Inspect `tool-calls.json` and `final.json` for actual tool use, `toolIssues`, `instructionIssues`, and usefulness, not just harness pass/fail.
+- Use `bun run agent:e2e` when changing MCP instructions, tool descriptions, or agent-facing tool behavior. This is a human/agent-driven qualitative eval, not a deterministic CI gate. Pick targeted workloads from `eval/agentic/README.md`; run both Claude and Codex for broad instruction changes when practical. Inspect `tool-calls.json` and `final.json` for actual tool use and the neutral answer/confidence, `metrics.json` for derived token/cost/duration/tool-call metrics, and `isolation-violations.json` for trace-validation failures, not just harness pass/fail. Treat usefulness or quality as reportable only when a later grading stage provides it.
 - Maintain smoke coverage when adding or changing user-facing tools/commands. Prefer structural UX assertions over brittle snapshots, and keep MCP `format: "json"` and CLI `--json` behavior aligned.
 - When changing GraphQL/API selections, add regression tests for over-fetch controls (for example `@include` variables, body omission, field lists, or query builders) and live-smoke the affected CLI/MCP surfaces when authenticated access is available.
 - Keep tests async and isolated
@@ -82,6 +94,7 @@ See `docs/guidelines/TESTING.md` for comprehensive patterns.
 - Root `skills/` and `AGENTS.md` are the only authored shared agent guidance. `CLAUDE.md` and `GEMINI.md` must remain symlinks to `AGENTS.md`.
 - Use the repository-internal `githits-plugin-maintenance` skill when changing skills, agent guidance, plugin/marketplace/extension manifests, MCP transport metadata, root release metadata, generator behavior, or agent-facing setup/auth behavior. It must remain under `.agents/skills/` and must not be published with the public root `skills/` tree.
 - Do not edit generated plugin assets directly. Change their canonical inputs, run `bun run plugins:generate`, inspect the diff, and run `bun run plugins:check`.
+- `packages/mcp/src/mcp/instructions.ts` owns the stable `buildMcpQuickStart()` guide. When it or the terminal guide section in `skills/githits-mcp/SKILL.md` changes, update both in the same PR; `src/skills-packaging.test.ts` exact-parity coverage is the contract. `buildLocalMcpQuickStart()` runtime appendices are excluded from the public skill copy. Behavior-dependent guide changes follow the public Agent Skill lifecycle.
 - `server.json` owns the canonical plugin keyword list used by generated manifests; keep `package.json` aligned with it.
 - All plugin and extension packages use hosted remote MCP. Direct `githits init` configuration retains local stdio except for Cursor, which is remote-only. Claude and Gemini direct setup remove legacy plugin or extension state before installing the user-scoped stdio server.
 

package/README.md

@@ -132,8 +132,9 @@ npx githits@latest docs read <page-id> --lines 20-80
 
 GitHits 0.10 adds two opt-in local tools for early dogfooding:
 
-- `resolve_target` / `githits resolve` turns a fuzzy or ambiguous package or
-  repository name into ranked canonical targets.
+- `resolve_target` / `githits resolve` turns a fuzzy or ambiguous package,
+  repository, or documentation-site name into grouped canonical targets with
+  related project identities kept together.
 - `code_diff` / `githits code diff` compares repository trees resolved from
   exact package versions or public GitHub refs.