[sentry-junior]

Follow-up to #2071, as BYK asked: remove asInvocationCommand(command) and the runtime startsWith("sentry ") rewriting of hints.

What changed

• asInvocationCommand is deleted. All 51 call sites are gone. ContextError, ResolutionError, buildValidationMessage and org-list now print the text they receive, with no rewriting.
• Stored commands are LazyCommand thunks. Any command kept in a constant or table is now () => sentry`…` and gets built when it is shown. That covers 98 module-level sentry … literals: USAGE_HINTs, commandPrefix, suggestion tables, SLUG_REDIRECT_COMMAND, and the issue, release, replay and dashboard argument builders. Every one of them was static text, so none needed runtime data.
• Recursion works by type, not by text. The sentry`…` tag inserts interpolated values as they are. A nested tag result already starts with the program name, and plain text that happens to start with "sentry " is no longer rewritten. That closes the gap where user input could get rewritten.
• Lint rule (sentry-cli/no-hardcoded-sentry-command):
  • The exemption for module-level whole-command literals is removed, so a bare 'sentry issue list …' constant is now flagged.
  • A module-level tag call is still flagged, and the message now points to the thunk form.
  • The asInvocationCommand handling is removed.
• A real bug is fixed. The issueTarget default sentry`issue` used to be evaluated at module load, before setAppIdentity. The new test/lib/app-identity-import.test.ts imports app.js and mcp-server-app.js and asserts that no hint gets built at import. I checked that it fails when that eager evaluation is put back.
• The packages/cli AGENTS.md note, the errors.ts JSDoc, the RuleTester cases and the app-identity tests are updated.

Design choice: thunks instead of a lazy object

I first tried having the tag return a lazy { toString() } command value. That produced 174 tsc errors across about 50 files, because every string slot would have to accept the new type. A value like that would also serialize as {} in JSON output. Thunks are type-checked, make the moment of evaluation explicit, and keep the literal tag text visible to check:doc-examples. Its tag scan still covers the text inside the thunks: 1649 examples in 669 files match the CLI.

One remaining prefix swap (open question)

The --help rendering of docs.examples still replaces a leading sentry with the running program, in a private exampleForRunningProgram in lib/command.ts. That data covers about 480 examples, and the generated docs and skill references share it, so they need the literal sentry name. The static alternative is to drop sentry from the docs.examples data and let each renderer add its own program name. That is a schema change across every command, so I left it out of this PR. I can do it as a follow-up if you want.

How it was tested

• Root pnpm run tsc: all packages pass.
• Root pnpm run lint: 0 warnings, 0 errors (includes ast-grep and the deps check).
• node --test scripts/cli-lint-rules.test.mjs: 116 pass.
• pnpm --filter sentry run check:doc-examples: 1649 examples match.
• pnpm --filter sentry run test: only the 5 known sandbox permission failures remain (completions EACCES/EPERM, resilient readonly, code-scanner unreadable files, upgrade lock). The 35 test failures from string fixtures were fixed by switching those fixtures to thunks, and those 4 files pass again (108 tests).

Refs #2071
Refs #2050

via @BYK.

--

View Session [Sentry]

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
sentry-local	Ready	Preview	Nov 7, 2026 10:55am UTC

[sentry-warden]

traceUsageHint under npx/mcp double-prefixes ambiguous-project errors

Building this hint with sentry makes it start with npx -y … or sentry-mcp, but commandNameFromUsageHint still strips only /^sentry\s+/, so ambiguous project-search errors rebuild suggestions like npx -y sentry npx -y sentry log list …. Strip via getInvocationCommand() (or pass a bare command name) before re-wrapping with sentry.

Evidence

• traceUsageHint is now () => sentry\log list …`and is passed intoresolveTraceOrg` (list.ts) as a full invocation string.
• commandNameFromUsageHint only strips /^sentry\s+/ (trace-target.ts), so under npx/mcp it returns e.g. npx -y sentry log list instead of log list.
• resolveTraceOrg → project-search → resolveOrgOnlyTarget/resolveProjectBoundTarget re-wraps that value with sentry\${commandName} …`` on ambiguous matches, producing a doubled program prefix.
• Default identity sentry still works; the failure needs npx (npm_command=exec / _npx) or setAppIdentity({ name: "sentry-mcp", … }) plus an ambiguous bare project + trace id.

Also found at 1 additional location

• packages/cli/src/commands/span/list.ts:186-186

_{Identified by Warden · code-review · D89-TDK}

[sentry-junior]

Confirmed and fixed in #2086. The trace resolvers now take the bare command name (log list, span list, span view, trace logs, trace view) and no longer derive it from the hint. commandNameFromUsageHint is deleted. A regression test checks that under sentry-mcp the program name appears only once.