Task 2.5: Select and apply built-in tools (Read, Write, Edit, Bash, Grep, Glob) effectively

• AGlob, with a pattern such as **/*parseInvoiceLine* to match the relevant files

  Incorrect

  Glob matches file paths, not file contents; callers live in files with unrelated names.

• BBash, running ls -R to list the repository tree and pick likely files to open

  Incorrect

  A directory listing shows names only and still leaves the agent guessing which files contain calls.

• CGrep, searching the contents of the repository's files for the function name

  Correct

  Grep searches inside files for patterns such as function names, which is exactly what finding callers needs.

• DRead, opening every file under src to look for the calls one file at a time

  Incorrect

  Reading everything up front is slow and fills the context with irrelevant code.

Use Grep for content searches such as callers, error messages and imports, then Read only the files the search turns up.

• ARead every file in the repository first so that no usage of the package can be missed

  Incorrect

  Reading everything up front wastes context and time on files that never touch the package.

• BList the names the wrapper exports, then Grep for each of those names across the codebase

  Correct

  Callers use the wrapper's names, so tracing usage means searching for every exported name, not only the original ones.

• CUse Grep to find entry points and imports, then Read those files and follow their imports outward

  Correct

  Searching first and reading along the import chain builds understanding incrementally without loading the whole tree.

• DUse Glob to search file contents for the package's original function names

  Incorrect

  Glob matches paths, not contents, and the original names miss calls made through the wrapper's new names.

• EGrep only for the package's import path, since every caller must import the package directly

  Incorrect

  Callers import the wrapper, not the package, so this search misses them.

Explore incrementally: search for entry points with Grep, read along the imports, and when a wrapper renames things, enumerate its exports and search for each name.

• AOne Edit call with replace_all set to true

  Correct

  replace_all replaces every occurrence, which is what this rename wants.

• BA Write call with the whole file retyped from memory

  Incorrect

  Retyping the file risks changing other code by accident.

• CA Grep call with the new name as the pattern

  Incorrect

  Grep searches; it does not change files.

• DA Glob call that renames matching text

  Incorrect

  Glob only matches file paths.

Edit requires a unique match unless replace_all is set; use replace_all when every occurrence in the file should change.

• AThe symbol must be quoted differently in the pattern

  Incorrect

  The pattern is fine; the file was never searched.

• BGrep skips gitignored files; pass the file's path

  Correct

  Grep respects .gitignore, and a direct path searches the ignored file.

• CGrep only searches files that have been read before

  Incorrect

  Read history does not limit Grep.

• DGrep cannot read generated files of any kind

  Incorrect

  It can, when given the path.

Grep skips gitignored files by default; Glob does not. To search an ignored file, pass its path to Grep explicitly.

• AGrep for the plain word interface and read every match to find the right ones

  Incorrect

  This works in principle but returns thousands of matches to read, which the precise pattern was meant to avoid.

• BWrap the pattern in quotes so that Grep treats it as literal text

  Incorrect

  Grep takes a regular expression; quoting the pattern does not change how its metacharacters are read.

• CSet multiline to true so that the pattern can match across the braces

  Incorrect

  Multiline matching lets a pattern span lines; it does not fix unescaped metacharacters.

• DEscape the metacharacters, for example interface\{\}, since Grep uses ripgrep's regex syntax

  Correct

  Grep is built on ripgrep, so braces and parentheses are regex syntax and need escaping to match literally.

Grep uses ripgrep's regular expressions, not literal text: characters such as { } ( ) need escaping, as in interface\{\}. A pattern ripgrep rejects returns an error with its diagnostic so the agent can correct the input and search again.

• ASwitch to Grep with an empty pattern to list all files

  Incorrect

  Grep searches contents; it is not a file lister.

• BRun the same pattern again to receive the next 100 paths

  Incorrect

  Glob does not page through results that way.

• CTreat the 100 paths as the complete list of TypeScript files

  Incorrect

  The truncation flag says the list is incomplete.

• DNarrow the pattern, for example to services/billing/**/*.ts

  Correct

  Glob results are capped at 100; a narrower pattern returns the files actually needed.

Glob returns up to 100 paths sorted by modification time and flags truncation, so narrow the pattern when the cap is hit.

• ABash is always preferred over built-in tools by design

  Incorrect

  The choice follows which tools the session has.

• BThe repository is too large for Grep and Glob

  Incorrect

  Size does not remove tools.

• CGrep and Glob are off by default there unless named in the options

  Correct

  On macOS, Linux and WSL the default set leaves them out, and Claude searches through Bash.

• DGrep and Glob only work in interactive Claude Code sessions

  Incorrect

  They are available in the SDK when included.

On macOS, Linux and WSL, Claude Code's default tool set omits Glob and Grep; naming them in tools or allowedTools brings them back.