[joeychilson]

Codex's reader names an exec script by the tools it calls, as exec: apply_patch, exec_command. It found them by every tools. in the script's text, strings included. So a script that only searches, await tools.exec_command({cmd: "rg -n tools.apply_patch src"}), was named exec: exec_command, apply_patch, as if it also patched files. sessions read, read_session and the handoff show that name, and #77 makes the handoff's reads_only decide by it, so with #77 such a script would still count as work.

The name now takes only calls: each tools.<name>( outside the script's strings. A string is one in " or `, to its closing quote, past \-escaped ones. That covers the shapes in the tests and the fixture (commands and patches in " strings) and patches in template literals, which agents::patched already expects. Not ': outside a string, a ' may be an apostrophe in a comment, and taking it for a string would hide every call after it. So a tool's call written inside a ' string, as 'tools.apply_patch(', is still named, as before. The decision is written beside the code.

The test agents::codex::tests::a_script_is_named_by_the_tools_it_calls_not_those_its_strings_name names three scripts: the one above, a patch in a template literal that adds a line calling tools.view_image(, and a command whose string holds an escaped \"tools.view_image(\" followed by a real tools.apply_patch( call. On origin/main it fails:

thread 'agents::codex::tests::a_script_is_named_by_the_tools_it_calls_not_those_its_strings_name' panicked at src/agents/codex.rs:1498:9:
assertion `left == right` failed
  left: ["exec: exec_command, apply_patch", "exec: apply_patch, view_image", "exec: exec_command, view_image, apply_patch"]
 right: ["exec: exec_command", "exec: apply_patch", "exec: exec_command, apply_patch"]

It passes with the fix. The existing test a_conversation_keeps_the_persons_words_and_names_what_a_script_did and the Codex snapshots (sessions-read-codex, sessions-read-failed) pass unchanged: no snapshot changed.

ingest::READERS stays as it is: the name is built only when a transcript is read, on demand, and transcripts aren't stored. CHANGELOG.md gets an entry under 0.1.5.

Checked: cargo fmt --check, cargo clippy --all-targets --locked -- -D warnings, cargo test --locked, TZ=America/Chicago cargo test --locked and RUSTDOCFLAGS="-D warnings" cargo doc --no-deps --locked all pass. Not run: the Swift tests, shellcheck, actionlint and cargo deny, since nothing they cover changed; the handoff cases, since the handoff's code didn't change. Not checked: how often real Codex scripts use ' strings or comments; I didn't look at real sessions.