TypeScript Type-Check with shell_agent Fallback
When verifying TypeScript compilation as part of a build or validation workflow, run_shell may fail with unknown errors due to shell environment issues, missing PATH entries, timeout constraints, or other transient problems. This skill describes a two-step pattern: attempt the check with run_shell first, then fall back to shell_agent if it fails.
When to Use
- After creating or modifying TypeScript source files and you need to confirm there are no type errors.
- As part of a CI-like verification step inside an autonomous workflow.
- Any time
run_shellreturns an unexpected or unclear error for atscinvocation.
Step-by-Step Pattern
Step 1 — Attempt with run_shell (fast path)
Try the direct command first. This is fast and gives immediate output when the environment is well-configured.
run_shell: npx tsc --noEmit
Evaluate the result:
- ✅ Exit code 0, no errors → Type-check passed. Done.
- ❌ Exit code non-zero with TypeScript diagnostics → Real type errors exist; surface them and fix.
- ⚠️ Unknown error, timeout, or no useful output → Proceed to Step 2.
Step 2 — Fall back to shell_agent (reliable path)
When run_shell fails in an ambiguous or environmental way, delegate to shell_agent with a natural-language task. shell_agent can autonomously:
- Locate the correct
npx/tscbinary - Retry on transient failures
- Navigate PATH or permission issues
- Return a structured summary of any type errors
Task string to pass to shell_agent:
Run `npx tsc --noEmit` in the project root and report all TypeScript type errors found. If there are no errors, confirm the compilation succeeded. If tsc is not available, attempt to install TypeScript locally with `npm install --save-dev typescript` first.
Evaluate the result:
- ✅ "No errors found" / exit 0 → Type-check passed.
- ❌ Type errors listed → Collect and fix the reported diagnostics, then re-run.
- ⚠️
tsconfig.jsonmissing → Ensure the file exists in the project root before re-running.
Decision Flowchart
run_shell("npx tsc --noEmit")
│
├─ success (exit 0) ──────────────────► ✅ PASS
│
├─ type errors reported ──────────────► ❌ Fix errors, retry
│
└─ unknown/env error ─────────────────► shell_agent(natural-language task)
│
├─ success ───► ✅ PASS
└─ errors ───► ❌ Fix & retry
Example Implementation
# Pseudo-code for an autonomous agent loop
result = run_shell("npx tsc --noEmit", timeout=60)
if result.exit_code == 0:
print("TypeScript check passed.")
elif result.exit_code != 0 and result.stderr contains "TS\d+":
# Real TypeScript diagnostics — surface and fix them
raise TypeScriptError(result.stderr)
else:
# Ambiguous failure — delegate to shell_agent
shell_agent(
task="Run `npx tsc --noEmit` in the project root and report all TypeScript "
"type errors found. If there are no errors, confirm compilation succeeded. "
"If tsc is not available, install it first with `npm install --save-dev typescript`."
)
Tips
- Always have a
tsconfig.json:tsc --noEmitrequires a valid config. If none exists, generate one first withnpx tsc --init. - Set a reasonable timeout: TypeScript compilation on large projects can take 30–120 seconds. Give
run_shellat leasttimeout=90. - Scope to changed files when possible: Use
--project tsconfig.jsonor a path glob to limit what tsc checks if only a subset of files changed. - Prefer
shell_agentfor first-time setups: If the project environment is freshly created and dependencies may not be fully installed, start withshell_agentdirectly to avoid a predictablerun_shellfailure.
Generalizing Beyond TypeScript
This same fallback pattern applies to any build/lint/type-check command that may be sensitive to shell environment:
| Tool | Primary run_shell command |
shell_agent fallback task |
|---|---|---|
| TypeScript | npx tsc --noEmit |
"Run npx tsc --noEmit and report errors" |
| ESLint | npx eslint src/ |
"Run npx eslint on the src directory and report lint errors" |
| Python mypy | python -m mypy src/ |
"Run mypy on the src directory and report type errors" |
| Go build | go build ./... |
"Run go build and report compilation errors" |
The core principle: run_shell for speed; shell_agent for resilience.
Converted and distributed by TomeVault — claim your Tome and manage your conversions.