docs: use installed ghcrawl binary in skill docs

Author: huntharo

skills/ghcrawl/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ghcrawl
 description: "Use the local ghcrawl CLI to inspect duplicate clusters and issue/PR summaries from the existing ghcrawl dataset, and refresh one repo only when the user explicitly asks. Use when a user wants to triage related issues or PRs, inspect semantic clusters, or run ghcrawl's staged refresh pipeline."
-allowed-tools: Bash(ghcrawl:*), Bash(pnpm:*), Read(*)
+allowed-tools: Bash(ghcrawl:*), Read(*)
 ---
 
 # ghcrawl
@@ -31,13 +31,9 @@ Also never run `close-thread` or `close-cluster` unless the user explicitly asks
 
 ## Command preference
 
-Prefer the installed `ghcrawl` bin.
+Use the installed `ghcrawl` bin directly.
 
-If `ghcrawl` is not on `PATH`, use:
-
-```bash
-npx ghcrawl cli ...
-```
+If `ghcrawl` is not on `PATH`, stop and tell the user the local `ghcrawl` binary is unavailable. Do not fall back to repo-local wrappers. If the user is actively maintaining `ghcrawl` itself and explicitly wants source-tree commands, use the maintainer workflow in `CONTRIBUTING.md`.
 
 Do not start by running `ghcrawl --help` or `<subcommand> --help`. The documented command surface in this skill and [references/protocol.md](references/protocol.md) is the default source of truth. Only use help output when the user explicitly asks about CLI syntax or you are actively maintaining ghcrawl itself.
 
@@ -95,12 +91,6 @@ Run:
 ghcrawl doctor --json
 ```
 
-If the bin is unavailable, fall back to:
-
-```bash
-pnpm --filter ghcrawl cli doctor --json
-```
-
 Only do this when:
 
 - the user explicitly wants an API-backed operation such as `refresh`, `sync`, `embed`, or `cluster`
@@ -115,12 +105,6 @@ If `doctor` is unhealthy but the user asked only for read-only inspection, say t
 
 ### 3. If the CLI is unavailable or misbehaving
 
-Use one supported fallback path before giving up:
-
-```bash
-pnpm --filter ghcrawl cli ...
-```
-
 If a documented `ghcrawl` command still fails, hangs, or returns unusable output through the supported CLI path, stop and report that to the user. Do not inspect tables, schema, or rows with `sqlite3`, `pragma`, or ad hoc SQL.
 
 ### 4. Refresh local data only when explicitly requested

skills/ghcrawl/references/protocol.md

@@ -167,26 +167,11 @@ Useful for semantic or keyword follow-up.
 
 Useful for inspecting nearest semantic matches for one thread.
 
-## Fallback invocation
-
-If `ghcrawl` is not installed globally:
-
-```bash
-pnpm --filter ghcrawl cli doctor --json
-pnpm --filter ghcrawl cli threads owner/repo --numbers 12345
-pnpm --filter ghcrawl cli threads owner/repo --numbers 42,43,44
-pnpm --filter ghcrawl cli threads owner/repo --numbers 42,43,44 --include-closed
-pnpm --filter ghcrawl cli author owner/repo --login lqquan
-pnpm --filter ghcrawl cli refresh owner/repo
-pnpm --filter ghcrawl cli clusters owner/repo --min-size 10 --limit 20 --sort recent
-pnpm --filter ghcrawl cli clusters owner/repo --min-size 10 --limit 20 --sort recent --include-closed
-pnpm --filter ghcrawl cli cluster-detail owner/repo --id 123 --member-limit 20 --body-chars 280
-pnpm --filter ghcrawl cli cluster-detail owner/repo --id 123 --member-limit 20 --body-chars 280 --include-closed
-pnpm --filter ghcrawl cli close-thread owner/repo --number 42
-pnpm --filter ghcrawl cli close-cluster owner/repo --id 123
-```
-
-If the supported CLI path still fails, hangs, or returns unusable output, stop and tell the user there is a ghcrawl CLI problem. Do not fall back to direct SQLite inspection.
+## CLI availability
+
+Use the installed `ghcrawl` binary directly.
+
+If `ghcrawl` is missing on `PATH`, stop and tell the user the local `ghcrawl` binary is unavailable. Do not fall back to project-local wrappers or direct SQLite inspection. If the user is actively maintaining `ghcrawl` itself and explicitly wants source-tree commands, use the maintainer workflow documented in `CONTRIBUTING.md`.
 
 ## Suggested analysis flow