tab: enforce deterministic model arg and improve tab-name extraction (opencode)#487
tab: enforce deterministic model arg and improve tab-name extraction (opencode)#487chr1syy wants to merge 1 commit intoRunMaestro:mainfrom
Conversation
📝 WalkthroughWalkthroughEnhanced the tab naming handler with stricter model resolution logic (sessionCustomModel > agent-config > default), comprehensive CLI argument sanitization and deduplication, improved SSH remote execution handling, and extensive debug logging throughout the flow. All changes contained within a single file. Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes 🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
src/main/ipc/handlers/tabNaming.ts (1)
261-262:⚠️ Potential issue | 🟡 MinorNon-adjacent includes check may produce false positives.
The condition
finalArgs.includes('--input-format') && finalArgs.includes('stream-json')could incorrectly return true if'stream-json'appears elsewhere in the args (e.g., as a value for a different flag). Consider checking adjacency:-const hasStreamJsonInput = - finalArgs.includes('--input-format') && finalArgs.includes('stream-json'); +const inputFormatIdx = finalArgs.indexOf('--input-format'); +const hasStreamJsonInput = + inputFormatIdx !== -1 && finalArgs[inputFormatIdx + 1] === 'stream-json';🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/main/ipc/handlers/tabNaming.ts` around lines 261 - 262, The current hasStreamJsonInput check can false-positive because it only checks finalArgs.includes('--input-format') && finalArgs.includes('stream-json'); instead, find the index of '--input-format' in finalArgs (e.g., const idx = finalArgs.indexOf('--input-format')) and set hasStreamJsonInput = idx !== -1 && finalArgs[idx + 1] === 'stream-json' so you verify adjacency; update the code around the hasStreamJsonInput variable in tabNaming.ts accordingly.
🧹 Nitpick comments (4)
src/main/ipc/handlers/tabNaming.ts (4)
135-152: Type casting suggests missing type definition forsessionCustomModel.The config parameter is cast to
anyto accesssessionCustomModel, but this property isn't defined in the config type (lines 83-91). If this property is expected, add it to the type definition for type safety.async (config: { userMessage: string; agentType: string; cwd: string; + sessionCustomModel?: string; sessionSshRemoteConfig?: { enabled: boolean; remoteId: string | null; workingDirOverride?: string; }; }): Promise<string | null> => {Then replace the
(config as any).sessionCustomModelcasts with direct property access.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/main/ipc/handlers/tabNaming.ts` around lines 135 - 152, The code uses (config as any).sessionCustomModel to read sessionCustomModel and bypasses typing; add sessionCustomModel?: string to the config type/interface used by the function that declares the config parameter so the property is properly typed, then remove the casts and access config.sessionCustomModel directly when setting resolvedModelId (the same pattern should be preserved for resolvedModelId, agentConfigValues.model and (agent as any).defaultModel — replace the (agent as any).defaultModel cast by adding defaultModel?: string to the Agent type/interface or using the existing Agent type so you can use agent.defaultModel directly).
391-402: MutatingresolvedModelIdaffects diagnostic logging.The mutation at line 401 changes
resolvedModelIdafter it was originally resolved. This variable is later used in logging (line 468), which will show the modified value rather than the original resolution. Consider using a separate variable for the final injected model.+ let finalModelForInjection = resolvedModelId; if ( - !resolvedModelId.includes('/') && + !finalModelForInjection.includes('/') && (agent as any).defaultModel && typeof (agent as any).defaultModel === 'string' && (agent as any).defaultModel.includes('/') ) { - resolvedModelId = (agent as any).defaultModel as string; + finalModelForInjection = (agent as any).defaultModel as string; }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/main/ipc/handlers/tabNaming.ts` around lines 391 - 402, The code currently mutates resolvedModelId (in the block that prefers agent.defaultModel when resolvedModelId lacks a provider prefix), which alters the value later used for diagnostic logging; instead, introduce a new variable (e.g., injectedModelId or cliModelId) to hold the model string that will be injected as a CLI flag and leave resolvedModelId unchanged for logs. Update the injection logic to assign to injectedModelId when you would have reassigned resolvedModelId and use injectedModelId for any CLI/config insertion while keeping resolvedModelId for later logging and diagnostics; reference resolvedModelId, (agent as any).defaultModel, and the injection site in tabNaming.ts when making the change.
161-171: Overly defensive try/catch around debug logging.The try/catch blocks wrapping
console.debugcalls appear throughout the file (lines 161-171, 190-201, 293-317, etc.).console.debugrarely throws, so this adds noise without meaningful protection.Consider a helper function if error isolation is truly needed:
const safeDebug = (message: string, data?: object) => { try { console.debug(message, data); } catch { /* swallow */ } };🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/main/ipc/handlers/tabNaming.ts` around lines 161 - 171, Remove the repetitive try/catch wrappers around console.debug in this module and replace them with a single safeDebug helper; create a top-level const safeDebug = (msg: string, data?: any) => { try { console.debug(msg, data); } catch { /* swallow */ } } and then change each protected console.debug call (the ones in tabNaming.ts that currently wrap sessionId/agentType/agentConfigModel/resolvedModelId and similar debug blocks) to call safeDebug(message, data) instead; this keeps error isolation but eliminates noisy duplicated try/catch blocks and centralizes behavior.
319-374: Redundant model flag processing.The deduplication logic (lines 323-357) extracts the last
--modelvalue and consolidates it, then lines 358-374 convert--model <value>to--model=<value>, and finally lines 376-424 strip all model tokens and re-inject the canonical one.Since lines 376-424 remove all model tokens anyway, the earlier deduplication and conversion steps are effectively discarded. Consider simplifying to just:
- Remove all model tokens
- Inject the canonical model
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/main/ipc/handlers/tabNaming.ts` around lines 319 - 374, The earlier two-step dedupe/convert logic around finalArgs (the inner try using lastModelVal and the rebuilt conversion into `--model=<value>`) is redundant because later code already strips model tokens and reinjects a canonical model; remove the whole deduplication block (the try { ... } that computes lastModelVal and the subsequent loop that builds rebuilt) and replace it with a single pass before the prompt separator: compute sepIndex from finalArgs, filter out any `--model` and `--model=...` tokens in finalArgs.slice(0, sepIndex), then if you have a canonical model id (use resolvedModelId or the previously determined canonical variable), push `--model=<canonical>` into the filtered args, and set finalArgs = [...filteredArgs, ...finalArgs.slice(sepIndex)]; keep the console.debug that logs sessionId and the canonical value.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Outside diff comments:
In `@src/main/ipc/handlers/tabNaming.ts`:
- Around line 261-262: The current hasStreamJsonInput check can false-positive
because it only checks finalArgs.includes('--input-format') &&
finalArgs.includes('stream-json'); instead, find the index of '--input-format'
in finalArgs (e.g., const idx = finalArgs.indexOf('--input-format')) and set
hasStreamJsonInput = idx !== -1 && finalArgs[idx + 1] === 'stream-json' so you
verify adjacency; update the code around the hasStreamJsonInput variable in
tabNaming.ts accordingly.
---
Nitpick comments:
In `@src/main/ipc/handlers/tabNaming.ts`:
- Around line 135-152: The code uses (config as any).sessionCustomModel to read
sessionCustomModel and bypasses typing; add sessionCustomModel?: string to the
config type/interface used by the function that declares the config parameter so
the property is properly typed, then remove the casts and access
config.sessionCustomModel directly when setting resolvedModelId (the same
pattern should be preserved for resolvedModelId, agentConfigValues.model and
(agent as any).defaultModel — replace the (agent as any).defaultModel cast by
adding defaultModel?: string to the Agent type/interface or using the existing
Agent type so you can use agent.defaultModel directly).
- Around line 391-402: The code currently mutates resolvedModelId (in the block
that prefers agent.defaultModel when resolvedModelId lacks a provider prefix),
which alters the value later used for diagnostic logging; instead, introduce a
new variable (e.g., injectedModelId or cliModelId) to hold the model string that
will be injected as a CLI flag and leave resolvedModelId unchanged for logs.
Update the injection logic to assign to injectedModelId when you would have
reassigned resolvedModelId and use injectedModelId for any CLI/config insertion
while keeping resolvedModelId for later logging and diagnostics; reference
resolvedModelId, (agent as any).defaultModel, and the injection site in
tabNaming.ts when making the change.
- Around line 161-171: Remove the repetitive try/catch wrappers around
console.debug in this module and replace them with a single safeDebug helper;
create a top-level const safeDebug = (msg: string, data?: any) => { try {
console.debug(msg, data); } catch { /* swallow */ } } and then change each
protected console.debug call (the ones in tabNaming.ts that currently wrap
sessionId/agentType/agentConfigModel/resolvedModelId and similar debug blocks)
to call safeDebug(message, data) instead; this keeps error isolation but
eliminates noisy duplicated try/catch blocks and centralizes behavior.
- Around line 319-374: The earlier two-step dedupe/convert logic around
finalArgs (the inner try using lastModelVal and the rebuilt conversion into
`--model=<value>`) is redundant because later code already strips model tokens
and reinjects a canonical model; remove the whole deduplication block (the try {
... } that computes lastModelVal and the subsequent loop that builds rebuilt)
and replace it with a single pass before the prompt separator: compute sepIndex
from finalArgs, filter out any `--model` and `--model=...` tokens in
finalArgs.slice(0, sepIndex), then if you have a canonical model id (use
resolvedModelId or the previously determined canonical variable), push
`--model=<canonical>` into the filtered args, and set finalArgs =
[...filteredArgs, ...finalArgs.slice(sepIndex)]; keep the console.debug that
logs sessionId and the canonical value.
Greptile SummaryThis PR refactors the tab-naming model resolution pipeline in Key issues found:
Confidence Score: 2/5
Important Files Changed
Flowchart%%{init: {'theme': 'neutral'}}%%
flowchart TD
A[Start: generateTabName] --> B[Fetch agentConfigsStore values]
B --> C{Resolve resolvedModelId}
C -->|config.sessionCustomModel?| D["sessionCustomModel branch\n⚠️ Dead code — field not in config type"]
C -->|agentConfigValues.model with '/'?| E[Use agent config model]
C -->|agent.defaultModel?| F[Use agent default model]
D --> G[Sanitize: remove trailing slashes]
E --> G
F --> G
G --> H[buildAgentArgs with modelId]
H --> I[applyAgentConfigOverrides]
I --> J[First sanitization: remove invalid --model flags]
J --> K{SSH enabled?}
K -->|Yes| L["buildSshCommand → finalArgs = SSH-wrapped args\n⚠️ Model injection happens AFTER this"]
K -->|No| M[Keep finalArgs as-is]
L --> N["Final safety sanitization\n(non-string removal)"]
M --> N
N --> O["Big sanitization block\n⚠️ Runs on SSH-wrapped args if SSH used"]
O --> P[Dedup --model flags]
P --> Q[Convert --model val to --model=val]
Q --> R[Strip ALL --model= and -m tokens]
R --> S{"resolvedModelId includes '/'?"}
S -->|Yes| T[Inject --model=resolvedModelId before '--']
S -->|No| U["⚠️ No model injected\n(all model flags silently dropped)"]
T --> V[Spawn process]
U --> V
Last reviewed commit: f958e5b |
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
| // Final safety sanitization: ensure args are all plain strings | ||
| try { | ||
| const nonStringItems = finalArgs.filter((a) => typeof a !== 'string'); | ||
| if (nonStringItems.length > 0) { | ||
| // eslint-disable-next-line no-console | ||
| console.debug('[TabNaming] Removing non-string args before spawn', { | ||
| sessionId, | ||
| removed: nonStringItems.map((i) => ({ typeof: typeof i, preview: String(i) })), | ||
| }); | ||
| finalArgs = finalArgs.filter((a) => typeof a === 'string'); | ||
| } | ||
|
|
||
| // Extract model arg value for debugging (if present) | ||
| const modelIndex = finalArgs.indexOf('--model'); | ||
| if (modelIndex !== -1 && finalArgs.length > modelIndex + 1) { | ||
| const modelVal = finalArgs[modelIndex + 1]; | ||
| // eslint-disable-next-line no-console | ||
| console.debug('[TabNaming] Final --model value', { | ||
| sessionId, | ||
| value: modelVal, | ||
| type: typeof modelVal, | ||
| }); | ||
| } | ||
| } catch (err) { | ||
| // swallow safety log errors | ||
| } | ||
|
|
||
| // Quote model values that contain slashes so they survive shell-based | ||
| // spawns (PowerShell can interpret unquoted tokens containing slashes). | ||
| try { | ||
| // Deduplicate --model flags and ensure exactly one is present before the prompt separator | ||
| try { | ||
| const sepIndex = | ||
| finalArgs.indexOf('--') >= 0 ? finalArgs.indexOf('--') : finalArgs.length; | ||
| let lastModelVal: string | undefined; | ||
| for (let i = 0; i < sepIndex; i++) { | ||
| if (finalArgs[i] === '--model' && finalArgs.length > i + 1) { | ||
| const cand = finalArgs[i + 1]; | ||
| if (typeof cand === 'string' && cand.trim()) { | ||
| lastModelVal = cand; | ||
| } | ||
| } | ||
| } | ||
|
|
||
| if (lastModelVal !== undefined) { | ||
| const newArgs: string[] = []; | ||
| for (let i = 0; i < sepIndex; i++) { | ||
| if (finalArgs[i] === '--model') { | ||
| i++; // skip value | ||
| continue; | ||
| } | ||
| newArgs.push(finalArgs[i]); | ||
| } | ||
| // Insert the single canonical model flag | ||
| newArgs.push('--model', lastModelVal); | ||
| // Append remaining args (including '--' and prompt) | ||
| finalArgs = [...newArgs, ...finalArgs.slice(sepIndex)]; | ||
| // eslint-disable-next-line no-console | ||
| console.debug('[TabNaming] Deduplicated --model flags', { | ||
| sessionId, | ||
| canonical: lastModelVal, | ||
| }); | ||
| } | ||
| } catch (err) { | ||
| // ignore dedupe failures | ||
| } | ||
| // Convert separate --model <value> pairs into a single --model=<value> | ||
| // token so shells don't split values. Then enforce a single canonical | ||
| // CLI model token derived from our resolvedModelId (if available). | ||
| const rebuilt: string[] = []; | ||
| for (let i = 0; i < finalArgs.length; i++) { | ||
| const a = finalArgs[i]; | ||
| if (a === '--model' && i + 1 < finalArgs.length) { | ||
| const raw = finalArgs[i + 1]; | ||
| const val = | ||
| typeof raw === 'string' ? raw.replace(/^['\"]|['\"]$/g, '') : String(raw); | ||
| rebuilt.push(`--model=${val}`); | ||
| i++; // skip the value | ||
| } else { | ||
| rebuilt.push(a); | ||
| } | ||
| } | ||
| finalArgs = rebuilt; | ||
|
|
||
| // Remove any existing model tokens (either --model=... or -m/value) | ||
| const withoutModel: string[] = []; | ||
| for (let i = 0; i < finalArgs.length; i++) { | ||
| const a = finalArgs[i]; | ||
| if (typeof a === 'string' && a.startsWith('--model')) { | ||
| // skip | ||
| continue; | ||
| } | ||
| if (a === '-m' && i + 1 < finalArgs.length) { | ||
| i++; // skip short form value | ||
| continue; | ||
| } | ||
| withoutModel.push(a); | ||
| } | ||
|
|
||
| // If we have a resolvedModelId (from session/agent/default), prefer inserting | ||
| // it explicitly as a CLI flag to avoid relying on OpenCode config/env. | ||
| if (resolvedModelId && typeof resolvedModelId === 'string') { | ||
| // If resolvedModelId doesn't look like provider/model, prefer agent.defaultModel | ||
| if ( | ||
| !resolvedModelId.includes('/') && | ||
| (agent as any).defaultModel && | ||
| typeof (agent as any).defaultModel === 'string' && | ||
| (agent as any).defaultModel.includes('/') | ||
| ) { | ||
| resolvedModelId = (agent as any).defaultModel as string; | ||
| } | ||
|
|
||
| if (resolvedModelId && resolvedModelId.includes('/')) { | ||
| const modelToken = `--model=${resolvedModelId}`; | ||
| // Insert before the argument separator `--` if present | ||
| const sep = withoutModel.indexOf('--'); | ||
| if (sep === -1) { | ||
| withoutModel.push(modelToken); | ||
| } else { | ||
| withoutModel.splice(sep, 0, modelToken); | ||
| } | ||
| // eslint-disable-next-line no-console | ||
| console.debug('[TabNaming] Injected canonical --model for spawn', { | ||
| sessionId, | ||
| model: resolvedModelId, | ||
| }); | ||
| } | ||
| } | ||
|
|
||
| finalArgs = withoutModel; | ||
| } catch (err) { | ||
| // swallow | ||
| } |
There was a problem hiding this comment.
Model injection runs after SSH command wrapping
The entire new sanitization and model-injection block (lines 292–424) executes after the SSH command is constructed (lines 244–290). When SSH is enabled, finalArgs at this point already contains the SSH-specific argument list (i.e., the remote agent command has been embedded as a single shell string or as SSH flags). The model-injection logic then scans those SSH args for -- separators and injects --model=<value> into the wrong position, corrupting the SSH command structure.
To fix this, the model injection must happen before buildSshCommand is called, so the model flag ends up inside the remote agent invocation, not in the SSH wrapper:
// ✅ Inject model flag here, before SSH wrapping
// ... (sanitize finalArgs)
if (config.sessionSshRemoteConfig?.enabled && ...) {
const sshCommand = await buildSshCommand(...)
command = sshCommand.command;
finalArgs = sshCommand.args;
}
// ❌ Do NOT modify finalArgs here for SSH sessions
| // Remove any existing model tokens (either --model=... or -m/value) | ||
| const withoutModel: string[] = []; | ||
| for (let i = 0; i < finalArgs.length; i++) { | ||
| const a = finalArgs[i]; | ||
| if (typeof a === 'string' && a.startsWith('--model')) { | ||
| // skip | ||
| continue; | ||
| } | ||
| if (a === '-m' && i + 1 < finalArgs.length) { | ||
| i++; // skip short form value | ||
| continue; | ||
| } | ||
| withoutModel.push(a); | ||
| } | ||
|
|
||
| // If we have a resolvedModelId (from session/agent/default), prefer inserting | ||
| // it explicitly as a CLI flag to avoid relying on OpenCode config/env. | ||
| if (resolvedModelId && typeof resolvedModelId === 'string') { | ||
| // If resolvedModelId doesn't look like provider/model, prefer agent.defaultModel | ||
| if ( | ||
| !resolvedModelId.includes('/') && | ||
| (agent as any).defaultModel && | ||
| typeof (agent as any).defaultModel === 'string' && | ||
| (agent as any).defaultModel.includes('/') | ||
| ) { | ||
| resolvedModelId = (agent as any).defaultModel as string; | ||
| } | ||
|
|
||
| if (resolvedModelId && resolvedModelId.includes('/')) { | ||
| const modelToken = `--model=${resolvedModelId}`; | ||
| // Insert before the argument separator `--` if present | ||
| const sep = withoutModel.indexOf('--'); | ||
| if (sep === -1) { | ||
| withoutModel.push(modelToken); | ||
| } else { | ||
| withoutModel.splice(sep, 0, modelToken); | ||
| } | ||
| // eslint-disable-next-line no-console | ||
| console.debug('[TabNaming] Injected canonical --model for spawn', { | ||
| sessionId, | ||
| model: resolvedModelId, | ||
| }); | ||
| } | ||
| } | ||
|
|
||
| finalArgs = withoutModel; |
There was a problem hiding this comment.
All model flags stripped but not re-injected when model lacks a provider prefix
The withoutModel loop unconditionally removes every --model=… and -m token from finalArgs (lines 376–389). The model is only re-injected at line 405 when resolvedModelId.includes('/'). If resolvedModelId was resolved from agent.defaultModel and that string has no / component (e.g. just a model short-name), neither the fallback at line 395 nor the injection at line 404 fires, and the spawn happens with no model argument at all — silently, because the outer try/catch swallows everything.
Trace for a model without a provider prefix:
resolvedModelId = "my-model"after initial resolution- Line 395:
!resolvedModelId.includes('/')istrue, but ifagent.defaultModelalso lacks/, the innerifis not entered andresolvedModelIdis unchanged - Line 404:
resolvedModelId.includes('/')isfalse→ no token injected finalArgs = withoutModel— all previous model flags stripped, nothing added back
The agent CLI then runs with whatever its built-in default model is, silently diverging from the configured value. The fix is to inject the model token regardless of whether it contains a /, or at minimum, avoid stripping model tokens that cannot be re-injected.
| let resolvedModelId: string | undefined; | ||
| if ( | ||
| typeof (config as any).sessionCustomModel === 'string' && | ||
| (config as any).sessionCustomModel.trim() | ||
| ) { | ||
| resolvedModelId = (config as any).sessionCustomModel.trim(); | ||
| } else if ( | ||
| agentConfigValues && | ||
| typeof agentConfigValues.model === 'string' && | ||
| agentConfigValues.model.trim() && | ||
| agentConfigValues.model.includes('/') | ||
| ) { | ||
| resolvedModelId = agentConfigValues.model.trim(); | ||
| } else if ( | ||
| (agent as any).defaultModel && | ||
| typeof (agent as any).defaultModel === 'string' | ||
| ) { | ||
| resolvedModelId = (agent as any).defaultModel as string; | ||
| } |
There was a problem hiding this comment.
sessionCustomModel branch is dead code — field is not in the handler config type
The IPC handler's config parameter (lines 83–92) is typed as:
config: {
userMessage: string;
agentType: string;
cwd: string;
sessionSshRemoteConfig?: { ... };
}sessionCustomModel is not part of this type. Accessing it via (config as any).sessionCustomModel will always evaluate to undefined, so the first branch of the model-resolution chain (lines 135–139) is never entered. The PR description lists "session override" as the highest-priority model source, but it will never be applied until sessionCustomModel is added to both the config type and the IPC caller.
| let resolvedModelId: string | undefined; | |
| if ( | |
| typeof (config as any).sessionCustomModel === 'string' && | |
| (config as any).sessionCustomModel.trim() | |
| ) { | |
| resolvedModelId = (config as any).sessionCustomModel.trim(); | |
| } else if ( | |
| agentConfigValues && | |
| typeof agentConfigValues.model === 'string' && | |
| agentConfigValues.model.trim() && | |
| agentConfigValues.model.includes('/') | |
| ) { | |
| resolvedModelId = agentConfigValues.model.trim(); | |
| } else if ( | |
| (agent as any).defaultModel && | |
| typeof (agent as any).defaultModel === 'string' | |
| ) { | |
| resolvedModelId = (agent as any).defaultModel as string; | |
| } | |
| let resolvedModelId: string | undefined; | |
| if ( | |
| typeof (config as any).sessionCustomModel === 'string' && | |
| (config as any).sessionCustomModel.trim() | |
| ) { | |
| resolvedModelId = (config as any).sessionCustomModel.trim(); | |
| } else if ( |
Consider adding sessionCustomModel?: string to the config interface and threading the value through from the caller, or removing this branch if it is not yet implemented.
| if (resolvedModelId && typeof resolvedModelId === 'string') { | ||
| // If resolvedModelId doesn't look like provider/model, prefer agent.defaultModel | ||
| if ( | ||
| !resolvedModelId.includes('/') && | ||
| (agent as any).defaultModel && | ||
| typeof (agent as any).defaultModel === 'string' && | ||
| (agent as any).defaultModel.includes('/') | ||
| ) { | ||
| resolvedModelId = (agent as any).defaultModel as string; | ||
| } |
There was a problem hiding this comment.
resolvedModelId mutation inside a swallowed try/catch causes misleading debug logs
resolvedModelId is mutated at line 401 inside the outer try/catch whose errors are silently swallowed (line 422–424). This variable is captured by reference in the onExit closure (line 464) and will reflect the post-mutation value when logged after process exit — even if the mutation happened in a code path that was not supposed to be the "winning" resolution.
More importantly, resolvedModelId was already used as sessionCustomModel in applyAgentConfigOverrides (line 184) with the original value. The late mutation only affects the final injection (line 405) and the debug logs, creating an inconsistency between what was passed to applyAgentConfigOverrides and what is ultimately injected/logged.
Consider using a separate variable (e.g. finalModelId) for this late fallback so the original resolvedModelId remains stable for logging purposes.
1 participant
This pull request makes significant improvements to the tab naming logic for agent-based sessions, focusing on more robust and predictable model resolution, enhanced argument sanitization, and improved debugging and logging. The changes ensure that the correct model is used for tab naming, prevent invalid or duplicate CLI arguments, and provide better diagnostics for troubleshooting tab naming quality.
Model Resolution and Argument Handling:
--modelflags, converting them to a single--model=<value>form, and ensuring only valid model values are passed to the agent CLI. Also, the canonical model flag is injected just before process spawn.Debugging and Logging Enhancements:
Tab Name Extraction Improvements:
Summary by CodeRabbit
Bug Fixes
Improvements