refactor(import): register transcript import per agent

[Stephen Belanger (Qard)]

Summary

Based on main now that #139 has merged. This is the last PR of the registration series.

Transcript import matched on ImportSource in seven places: roots, session id, filename match, envelope construction, tail state construction (twice), and polling. The generic TranscriptTail also had about 60 lines of Cursor-only snapshot and rewrite handling inline.

Each importable agent now implements a capability trait and supplies its own tail:

pub(crate) trait TranscriptImport: Agent {
    fn discover(&self, home: &Path) -> anyhow::Result<Vec<PathBuf>>;
    fn find(&self, home: &Path, session_id: &str) -> anyhow::Result<PathBuf>;
    fn tail(&self) -> Box<dyn TailSession>;
}

pub(crate) trait TailSession: Send {
    fn envelopes(&mut self, path: &Path, records: &IncrementalRecords) -> anyhow::Result<Vec<Envelope>>;
    fn poll(&mut self, events: Vec<Envelope>, len: u64, finalize: bool) -> anyhow::Result<Vec<Envelope>>;
    fn keeps_translator_on_rewrite(&self) -> bool { false }
    fn rewritten(&mut self, previous: &[Value], current: &[Value]) {}
    fn tolerates_incomplete_final_record(&self) -> bool { false }
}

• Registration. The Codex, Claude Code, Cursor and Antigravity agents register TranscriptImport.
• Agent tails. Each agent's existing Tail struct implements TailSession.
• Cursor tail. Cursor's tail owns the private transcript snapshot. It reports that it keeps its translator when the transcript is rewritten and that it tolerates an incomplete final record. Its rewritten hook computes which records survived a rewrite.
• Generic tailer. TranscriptTail holds a Box<dyn TailSession> and no longer mentions any agent.

Discovery is internal to each agent. Callers only need discover, find and tail, so each agent's search roots and filename rules are private to its module. The shared directory walk, session grouping and error messages live in a small TranscriptLayout (roots, a session-id rule and a filename rule) that each agent builds privately. Tests exercise each agent's layout with temporary roots, so they stay independent of the CODEX_HOME and CLAUDE_CONFIG_DIR overrides.

Behaviour

None. Search paths, matching and error messages are unchanged. Cursor's snapshot is still recreated after a rewrite and receives only the records that changed. Other agents still get a fresh translator after a rewrite.

Testing

• cargo clippy --all-features --all-targets -D warnings is clean.
• cargo test --all-features --no-fail-fast: everything passes, including the Cursor attach/rewrite import tests, except trace_runtime::tests::doctor_reports_a_daemon_that_{cannot_authenticate_the_route,resolves_other_credentials}. Those fail the same way on main in my local environment.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-06T12:49:52.469541Z	fa4201a	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review" or "Codex (@codex) security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.