fix(core): propagate SHELL→TERMINAL_SHELL rename into direct-shell allowlists (follow-up to #12021)

[lalalune]

The regression

PR #12021 renamed the terminal action SHELL → TERMINAL_SHELL (packages/agent/src/actions/terminal.ts) but did not propagate the new name into two hardcoded shell-name allowlists in packages/core/src/services/message.ts:

• WEAK_DIRECT_REPLY_OVERRIDE_ACTIONS
• SHELL_DIRECT_ACTIONS

Both sets are built by mapping their raw entries through normalizeActionIdentifier (.toUpperCase().replace(/_/g, "")), so "TERMINAL_SHELL" normalizes to "TERMINALSHELL". Neither set listed it, so the renamed action was never a member.

Lean-chat repro

In a lean-chat deployment (ELIZA_PLUGIN_SET=lean-chat, no plugin-coding-tools), TERMINAL_SHELL is the only shell action. For a message like run df -h on this VPS:

• inferDirectCurrentRequestCandidateActions still resolves TERMINAL_SHELL (via its RUN_IN_TERMINAL simile), so directCandidateActions = ["TERMINAL_SHELL"].
• but shouldPreferDirectCurrentCandidateActions then does SHELL_DIRECT_ACTIONS.has("TERMINALSHELL") → false (and the candidateActions.every(...) weak-override check also fails), so the turn is not promoted to the direct shell path and silently falls through to the general planner.

The fix

• Add "TERMINAL_SHELL" to WEAK_DIRECT_REPLY_OVERRIDE_ACTIONS and SHELL_DIRECT_ACTIONS.
• Also add it to the two findAvailableActionName shell-name lookup lists (in message.ts inferAckIntentCandidateActions and message/direct-action-heuristics.ts). These already resolved via simile; adding the canonical name is defensive and placed after "SHELL" so mixed deployments (coding-tools SHELL present) keep their existing priority.
• Other candidates checked and left alone: prompt-compaction.ts already lists both SHELL and TERMINAL_SHELL; planner-loop.ts name === "SHELL" is the coding-tools trajectory summary (unrelated); the many "SHELL" env-var/$SHELL references are unrelated.

Test

Added a regression test in message-routing-live-regression.test.ts that drives the real seam with a lean-chat fixture (only TERMINAL_SHELL + REPLY): it asserts inference resolves ["TERMINAL_SHELL"] and shouldPreferDirectCurrentCandidateActions returns true. Verified it fails pre-fix (returned false) and passes after.

Verification

• bun run --cwd packages/core typecheck → clean
• message-routing-live-regression.test.ts → 41 passed; direct-action-heuristics.test.ts → 5 passed

[greptile-apps]

Your trial has ended. Reactivate Greptile to resume code reviews.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 33e99fbb-fbbe-4499-a511-a604f0700d51

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/12021-terminal-shell-direct-allowlist

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[claude]

Claude encountered an error —— View job

---

I'll analyze this and get back to you.