buddy-agent-harness 1.0.1 → 1.2.0

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
 	"name": "buddy-agent-harness",
-	"version": "1.0.1",
+	"version": "1.2.0",
 	"description": "Initialize standards-based agent configuration across agent harnesses",
 	"author": {
 		"name": "repobuddy"

package/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
 	"name": "buddy-agent-harness",
-	"version": "1.0.1",
+	"version": "1.2.0",
 	"description": "Initialize standards-based agent configuration across agent harnesses",
 	"author": {
 		"name": "repobuddy"

package/.cursor-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
 	"name": "buddy-agent-harness",
-	"version": "1.0.1",
+	"version": "1.2.0",
 	"description": "Initialize standards-based agent configuration across agent harnesses",
 	"author": {
 		"name": "repobuddy"

package/CHANGELOG.md

@@ -1,5 +1,49 @@
 # buddy-agent-harness
 
+## 1.2.0
+
+### Minor Changes
+
+- 3d86bc2: Account for the git worktrees Claude Code creates under `.claude/worktrees/`, inside the repository.
+  
+  The `init-buddy-agent-harness` and `enhance` skills now offer two settings, each on its own and written only on approval: ignoring `.claude/worktrees/` in `.gitignore` and in each tree-scanning tool's config the repository has (Biome, ESLint, Prettier, knip, tsconfig), and setting `worktree.baseRef` to `"head"` in `.claude/settings.json`, with the trade-off explained first. Each worktree is a whole checkout, so Biome in the main checkout stops on its nested root configuration, and without `"head"` an agent asked to build on a feature branch starts from the default branch.
+  
+  `doctor` reports what is missing in a repository with a `.claude/` directory: `worktrees-untracked` for the `.gitignore` rule, `worktrees-scanned` for a `biome.json` that does not exclude the directory, and `worktree-base-unset` for a `.claude/settings.json` that leaves `worktree.baseRef` unset. An explicit `"fresh"` is the owner's choice and is not reported.
+- 8373121: The `enhance` skill now offers a fourth addition, `## References as links`. It tells the agent to write each issue, pull request, or commit it mentions as a Markdown link to its page rather than a bare `#222`, which is a link only on GitHub itself and names the wrong repository when the agent works across several.
+  
+  It is gated like the other three. The skill offers it only where the merged instructions do not already cover the subject, recommends the owner's global instruction file over the repository's `AGENTS.md`, and writes nothing without approval. Its history file starts empty, so a section already in a file is compared against the current text, and no harness scores this wording yet, so where the skill cannot tell whose a section is, it offers two answers rather than three.
+
+### Patch Changes
+
+- ce2b780: The `doctor-buddy-agent-harness` skill now lists the outcomes it is done when, so its evals can check them.
+- 8588ad8: The shipped skill descriptions (`doctor-buddy-agent-harness`, `enhance`, `init-buddy-agent-harness`, `repair`) now open by saying what the skill does instead of "Use this skill when". Every trigger phrase is kept.
+- 4c76524: The `repair` skill no longer says that only the instruction, configuration, non-standard, and agent-worktree families name a skill in `help`. Four MCP findings name `repair` too: `mcp-unprojected`, `mcp-undeclared`, `mcp-diverged-target`, and `mcp-diverged-golden`. Its README now lists the MCP drift it corrects.
+
+## 1.1.0
+
+### Minor Changes
+
+- 211a1f6: Upgrade to clibuilder 12.2 and follow its AXI conventions.
+  
+  - A failure is a coded error on stdout, in the format asked for, with a `help` line, instead of `error:` prose on stderr: `command-failed` (exit 1, any unexpected failure), `golden-set-unreadable`, `accept-rejected` (with the refusal `reasons`), `governance-not-found` and `unsupported-harness` (exit 2, a usage error). The governance deprecation warning stays on stderr.
+  - `mcp project` and `mcp reconcile` with no golden set answer with `empty: 0 servers — no golden set at …` and a `help` line, instead of a report of placeholder rows.
+  - A bare `buddy-agent-harness` answers with the `doctor` report under a `bin:` and `description:` header, instead of a usage error.
+  - Every command has `examples` in its `--help`.
+  
+  Migration: a script that read the failure message from stderr should read `error` and `code` on stdout, or use the exit code.
+- 16de22d: Upgrade to clibuilder 12.3 and cyber-agentkit 2.2.
+  
+  - Every `reference` subcommand now carries `examples` in its `--help`, each starting with `buddy-agent-harness reference`.
+  - `governance show`'s `governance-not-found` help is checked against the command tree, and names the command line the group is reached by: `buddy-agent-harness reference list` standalone, `repobuddy agent-harness reference list` when mounted under `repobuddy`. It no longer ends with a period.
+  - From cyber-agentkit 2.2, `reference search` shows the best 10 matches unless `--full` is given, and `reference customize` and `reference implementations` accept a `<prefix>.*` family. `reference migrate` is documented on the `reference` page.
+
+### Patch Changes
+
+- 30d912b: The `enhance` skill's `## List identifiers` section is now two sentences. It labels each option, question, finding, or proposal put to the user with a kind letter and a number, and it leaves the points of a plain explanation unlabelled, which the previous wording labelled. Labels still hold through a revision.
+  
+  A repository holding the previous wording is offered the new one as a replacement.
+- 05ff513: `mcp project` and `mcp reconcile` now give their next-step command through clibuilder's `ref()`, so the command line is checked against the CLI's command tree. With no golden set, `mcp project` answers two `help` lines: what to create, then ``Run `buddy-agent-harness mcp project` to preview the projection``. A rejected `mcp reconcile --accept` says ``Run `buddy-agent-harness mcp reconcile` to list the fields that can be approved``.
+
 ## 1.0.1
 
 ### Patch Changes