buddy-agent-harness 1.0.1 → 1.2.0

package/skills/repair/README.md

@@ -4,7 +4,7 @@ Correct agent configuration a repository already has that is wrong or outdated.
 
 ## What it does
 
-`init` consolidates what a repository has, `enhance` offers what it is missing, and `doctor` reports what is wrong. Configuration that is present but **wrong** falls between all three, and that is what this skill takes: a harness name the registry retired, a git-ignored bridge, an `AGENTS.local.md` nothing reads, a skill whose frontmatter makes a harness skip it.
+`init` consolidates what a repository has, `enhance` offers what it is missing, and `doctor` reports what is wrong. Configuration that is present but **wrong** falls between all three, and that is what this skill takes: a harness name the registry retired, a git-ignored bridge, an `AGENTS.local.md` nothing reads, a skill whose frontmatter makes a harness skip it, an MCP server that differs between the golden set and a harness on the side that is plainly ahead.
 
 Each correction is offered with the file as it stands and as it would read, and written only on approval.
 
@@ -12,7 +12,7 @@ Each correction is offered with the file as it stands and as it would read, and
 
 `doctor` stays read-only. Its whole safety property is that it never writes, which is what makes it safe to run from a session-start hook — a repair flag would forfeit that for every caller. Its **bridge** findings are all repairable with existing `init` flags, so a flag there would duplicate `init` rather than add anything; its **configuration** findings need judgment about files you wrote, which is a skill's job and not a flag's.
 
-That split also defines this skill's remit. `repair` owns what `init` deliberately refuses. Every bridge finding — including an instruction bridge that was never completed — is reported and handed to `init`, which is what writes bridges in the first place.
+That split also defines this skill's remit. `repair` owns what `init` deliberately refuses, and the four MCP findings where one side is plainly the one to copy from: `mcp project` or `mcp reconcile` works out the change, and this skill offers it. Every bridge finding — including an instruction bridge that was never completed — is reported and handed to `init`, which is what writes bridges in the first place.
 
 ## The line it does not cross
 

package/skills/repair/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: repair
-description: Use this skill when a repository's agent configuration is present but wrong or outdated — a retired harness name, a git-ignored bridge, an AGENTS.local.md nothing reads, a skill a harness refuses to load, or an MCP server that differs between the golden set and a harness, on the side that is plainly ahead. Runs doctor to find them, then offers each correction with its before and after and writes only what is approved.
+description: Repairs a repository's agent configuration when it is present but wrong or outdated — a retired harness name, a git-ignored bridge, an AGENTS.local.md nothing reads, a skill a harness refuses to load, or an MCP server that differs between the golden set and a harness, on the side that is plainly ahead. Runs doctor to find them, then offers each correction with its before and after and writes only what is approved.
 ---
 
 # Harness Repair
@@ -17,7 +17,7 @@ Each of the other three refuses this work deliberately. `init` has to run anywhe
 node scripts/doctor.mjs
 ```
 
-That path is relative to this skill's own directory: `scripts/doctor.mjs` is a self-contained bundle shipped inside this skill folder through the npm package, so it runs against the current working directory with nothing downloaded and no `node_modules` needed. Fall back to `npx -y buddy-agent-harness@1.0.1 doctor` when `scripts/doctor.mjs` is missing or cannot be run — the case for a skill installed from git rather than from the npm package, which does not carry the bundle.
+That path is relative to this skill's own directory: `scripts/doctor.mjs` is a self-contained bundle shipped inside this skill folder through the npm package, so it runs against the current working directory with nothing downloaded and no `node_modules` needed. Fall back to `npx -y buddy-agent-harness@1.2.0 doctor` when `scripts/doctor.mjs` is missing or cannot be run — the case for a skill installed from git rather than from the npm package, which does not carry the bundle.
 
 The default output is TOON, which is what you parse. Each entry in `findings` carries three fields: `problem` (the finding's name), `path` (what it is about), and `detail` (what is wrong, in prose). The repairs are not on the row — they are lifted into `help`, which follows `findings` in order and collapses two entries only where the repair is word-for-word identical. Read a finding's repair off the matching position and confirm it by the `path` it names; where two findings sit at one path, what each repair says to do is what tells them apart.
 
@@ -31,7 +31,7 @@ If `findings` reports zero problems, say so and stop. There is nothing to repair
 
 That rule holds for every finding `doctor` reports today and for the ones it reports next year. A fault this skill has no correction for is not one it can repair, so what `classes.md` covers and what this skill owns are the same fact — and a finding added tomorrow gets reported rather than improvised on.
 
-**Do not read ownership off a skill name in `help`.** Most repairs name no skill — some are a shell invocation, some a `git` command, some a reconciliation by hand — and only the instruction and configuration families name one at all. An owner is not something the report can always be asked for.
+**Do not read ownership off a skill name in `help`.** Most repairs name no skill — some are a shell invocation, some a `git` command, some a reconciliation by hand — and only the instruction, configuration, non-standard, and agent-worktree families, and four MCP problems, name one at all. An owner is not something the report can always be asked for.
 
 - **Findings that are yours.** Look the `problem` up in `references/classes.md`, which carries the correction and the stopping point for each.
 - **Findings that are not yours.** Report each one and pass on the repair `doctor` states for it, unchanged. **Who to hand it to is read off that repair, and the question is whether it names `init`.** Some name the `/buddy-agent-harness:init-buddy-agent-harness` (pnpm install: `pnpm-buddy-agent-harness-init-buddy-agent-harness`) skill and some name a `buddy-agent-harness init` command line; both mean the same thing — the finding is `init`'s, because `init` is what writes a bridge and what consolidates an instruction file in the first place, and deciding what to preserve while it does is judgment this skill's corrections never carry. Hand it to the **skill** and never run the command yourself: rebuilding a projection can relocate skills a user wrote, and on the Windows case the naive repair is to recreate the link, the operation that already failed on that machine. A repair naming `init` in **neither** form — a `git` invocation, a reconciliation by hand, an MCP finding with no section in `classes.md` — is work for a person; say that rather than inferring an owner, and never rebuild over a two-sided divergence, which discards whichever side holds the newer edit.