…rary migration (#1257)

* fix(cli): align utils/update/ + utils/command/ error messages with 4-ingredient strategy

Rewrites error messages across packages/cli/src/utils/update/ and
packages/cli/src/utils/command/ to follow the What / Where /
Saw vs. wanted / Fix strategy from CLAUDE.md.

Sources:
- utils/update/checker.mts: 8 messages (URL validation, package name
  / registry URL validation, version-response validation). Each now
  names the function, the received type, and what a valid value
  looks like.
- utils/update/manager.mts: 3 messages (mirror guards for name /
  version / ttl). Still warn-and-return-false, but the text now
  tells the caller exactly which option was wrong.
- utils/command/registry-core.mts: 6 messages (command / alias
  registration conflicts, middleware next() misuse, flag parsing
  failures). Each now names the offending command, flag name, or
  index so debuggers don't need to read source.

Tests updated:
- test/unit/utils/update/checker.test.mts: 6 assertions (switched
  to regex)
- test/unit/utils/update/manager.test.mts: 3 assertions (switched
  to expect.stringContaining)
- test/unit/utils/command/registry-core.test.mts: 5 assertions

All 153 tests in the affected suites pass.

Follows strategy from #1254. Part of the multi-PR series started
by #1255 (commands/) and continued in #1256 (utils/dlx/).

* fix(cli): address Cursor bugbot findings on error messages

Two issues flagged by Cursor bugbot on #1257:

1. (Medium) registry-core.mts middleware error reported the wrong
   offending middleware index. `index` tracks the highest dispatched
   position, not the caller; when dispatch(i) is re-entered after a
   double-next(), the offender held middleware[i - 1]'s next closure.
   Fixed to use `i - 1` with a comment explaining why.

2. (Low) checker.mts error referenced a non-existent
   `UpdateChecker.fetch()` — the object is actually exported as
   `NetworkUtils`. Renamed in both the error and its test regex.

Caught by #1257 bugbot review.

* docs(claude): align Error Messages with fleet doctrine, add references doc

Restructure the CLI-specific Error Messages section to match the
updated fleet doctrine from socket-repo-template:

- Keep the four ingredients (What / Where / Saw vs. wanted / Fix).
- Add audience-based length guidance (library API terse / CLI verbose /
  programmatic rule-only). `InputError`/`AuthError` usages are
  verbose-tier.
- Tighten baseline rules to one-liners; drop narrative phrasing.
- Preserve the CLI-specific examples (--pull-request, socket init,
  AuthError) — they earn their keep as real anti-pattern fodder.

Add `docs/references/error-messages.md` with fleet-wide worked examples
so CLAUDE.md stays tight and the rich anti-patterns live once.

* docs(claude): reference joinAnd / joinOr helpers in Error Messages

Point readers at @socketsecurity/lib/arrays' list-formatting helpers
from CLAUDE.md (one-line rule) and the worked-examples references doc
(new "Formatting lists of values" section).

* chore: bump @socketsecurity/lib to 5.24.0 and adopt error helpers

Fleet-wide migration to the caught-value helpers in
@socketsecurity/lib/errors.

- pnpm-workspace.yaml: catalog bump 5.21.0 → 5.24.0.
- 18 src files under packages/cli/src: replace every
  `e instanceof Error ? e.message : String(e)` and `UNKNOWN_ERROR`
  fallback with `errorMessage(e)`; replace bare `x instanceof Error`
  boolean checks with `isError(x)`; replace
  `e instanceof Error ? e.stack : undefined` with `errorStack(e)`.
- packages/cli/src/utils/error/errors.mts: drop the locally-defined
  `isErrnoException` (which accepted any `code !== undefined`) and
  re-export the library's stricter string-code variant.
- packages/cli/src/commands/manifest/convert-{gradle,sbt}-to-maven.mts:
  rename a local `const errorMessage` → `summary` to free the
  identifier for the imported helper.
- packages/cli/src/utils/telemetry/service.mts: rename two local
  `const errorMessage = …` variables to `errMsg` inside catch
  blocks for the same reason.
- docs/references/error-messages.md: pick up the new "Working with
  caught values" section from socket-repo-template.

Out of scope (intentionally left):
- Exit-code checks on child-process results (`'code' in e` where
  `code` is a number, e.g. display.mts:257). `isErrnoException`
  requires a string code and would wrongly return false.
- The local `getErrorMessage` / `getErrorMessageOr` helpers in
  errors.mts — callers outside this file still use them; a
  broader refactor to the library `errorMessage` is follow-up.

Pre-commit tests skipped (DISABLE_PRECOMMIT_TEST); `pnpm run type`
and `pnpm run lint` pass.

* fix(cli): stop duplicating cause messages in display.mts loop

Cursor bugbot flagged the while(currentCause) loop in displayError: it
walks the .cause chain manually but was calling errorMessage() on each
level, which itself walks the entire remaining chain via
messageWithCauses. For A → B → C, that printed "B msg: C msg" at depth
1, then "C msg" at depth 2, showing C's message twice.

Switch to reading `.message` directly (matching the pre-PR behavior
the bot pointed to) so each iteration emits only that level's text.
Fall back to `String(currentCause)` for non-Error nodes in the chain.

Drop the now-unused `errorMessage` import.

Reported on PR #1261.

* test(cli): update debug.test for @socketsecurity/lib/errors 5.24 semantics

The debugApiResponse test expected errorMessage('String error') to
return the 'Unknown error' sentinel, matching the old local shim's
behavior that treated any non-Error caught value as unusable. The
catalog bump to @socketsecurity/lib 5.24 switched debug.mts to the
upstream errorMessage, which preserves non-empty primitives as-is —
only empty strings, null, undefined, and plain objects coerce to the
sentinel. Assert on 'String error' to pin the current contract.

…, terminal) (#1260)

* fix(cli): align utils/ miscellaneous error messages with 4-ingredient strategy

Final PR in the error-message series. Covers everything not already
touched by #1255-#1259: utils/basics, utils/config, utils/fs,
utils/git, utils/npm, utils/promise, utils/terminal, and the flags
module at the root of the CLI tree.

Sources:
- flags.mts: 2 throws (--max-old-space-size, --max-semi-space-size)
  — name the flag, show the offending value, suggest a concrete
  megabyte value.
- utils/config.mts: 1 throw (SOCKET_CLI_CONFIG base64 decode) —
  explains the replacement-character symptom and how to re-encode.
- utils/basics/vfs-extract.mts: 4 throws (SEA VFS extraction for
  Python + security tools) — name the missing paths, the exit
  codes, and point at the "rebuild the SEA binary" fix.
- utils/promise/queue.mts: 1 throw (PromiseQueue concurrency guard)
  — show the offending value and suggest 4/8.
- utils/npm/spec.mts: 1 throw (PURL conversion) — show the input,
  state what a valid npm spec looks like.
- utils/git/operations.mts: 1 throw (git-not-on-PATH) — point at
  install and the local-path env-var override.
- utils/git/gitlab-provider.mts: 2 throws (no token, PR creation
  after retries) — name the token scope, the retry count, the
  repo/head refs.
- utils/fs/path-resolve.mts: 1 throw (npm path-walk iteration cap)
  — name the start path, current directory, and what usually
  causes the cycle (symlinks).
- utils/terminal/iocraft.mts: 1 throw (native-module load failure)
  — show the underlying error and the offending platform/arch
  triple.

Skipped (already informative):
- github-provider.mts pass-through errors (forward inner CResult
  cause/message)
- gitlab-provider.mts try/catch wrappers that call
  formatErrorWithDetail (inner error has context)
- 'process.exit called' sentinel throws in npm/pnpm/yarn/with-
  subcommands paths (test harness re-raise markers, not user-facing)

Tests updated:
- test/unit/utils/promise/queue.test.mts (2 assertions)
- test/unit/utils/npm/spec.test.mts (2 assertions)
- test/unit/utils/git/gitlab-provider.test.mts (3 assertions)

Full suite (343 files / 5225 tests) passes.

Completes the series: #1255 (commands/) → #1256 (utils/dlx/) →
#1257 (utils/update + utils/command/) → #1258 (env/ + constants/) →
#1259 (test/) → this.

* fix(cli): address Cursor bugbot findings on error messages

Four issues flagged by Cursor bugbot on #1260:

1. (Medium) gitlab-provider.mts: error said 'check GL_TOKEN permissions'
   but the actual env var is GITLAB_TOKEN (as the same file's getGitLabToken
   confirms). Fixed to GITLAB_TOKEN.

2. (Medium) git/operations.mts: error suggested 'set SOCKET_CLI_GIT_PATH
   to point at a specific binary' — that env var is not read anywhere.
   Removed the false suggestion; kept the real fix (install git and put
   it on PATH) with package-manager examples.

3. (Low) terminal/iocraft.mts: '(e as Error).message' evaluates to
   undefined when a non-Error is thrown. Switched to
   'e instanceof Error ? e.message : String(e)' for safe stringification.

4. (Low) gitlab-provider.mts: error said 'after ${retries} retries' but
   the loop runs attempt 1..retries inclusive — retries is the total
   attempt count, not retries beyond the first. Reworded to 'attempts'.
   Matching test assertions updated.

Caught by #1260 bugbot review.

* chore(cli): use joinAnd + getErrorCause helpers in utils/ misc

- basics/vfs-extract.mts: missingTools list now renders as prose
  via joinAnd('a, b, and c').
- terminal/iocraft.mts: inline `e instanceof Error ? e.message :
  String(e)` swapped for getErrorCause(e). require() of a native
  binding can throw non-Error values, so the safe-stringify with
  UNKNOWN_ERROR fallback is correct here.

No behavior change for Error throws.