cwilliams5/review-option-interaction

Enter planning mode. Deep-research all interactions between installation options, launch modes, admin state, and user journeys. Think carefully about every path a user could take. Use maximum parallelism — spawn an agent team if desired.

The Problem

The app has many options, features, and launch modes that interact combinatorially. Past reviews have repeatedly found edge-case bugs where one option conflicts with another or a UAC refusal leaves the system in a half-applied state. The combinatorial space is large and keeps growing.

Step 1 — Inventory the Moving Parts

Before testing combinations, build a current inventory by reading the code. These categories have historically caused interactions, but the code is the source of truth — discover what exists now, don't assume this list is complete:

User-facing options:

• Install to Program Files
• Run at startup (scheduled task)
• Run as admin
• Add to Start Menu
• Check for updates
• First-launch wizard

Features with interaction surfaces:

• Multi-process architecture (launcher + gui + pump)
• Auto-update (rename running exe, replace, relaunch)
• Duplicate version detection (already running → compare versions → offer restart/upgrade)
• Wizard skip / duplicate version skip flags
• Various launch flags (--gui-only, --config, --wizard-continue, --enable-admin-task, etc.)
• Menu items / dashboard actions that trigger installs or config changes

Confounding factors (things users do that the code must handle):

• Rename the exe (e.g., "AltTabby new v4.exe") — does every path handle non-standard names?
• Rename the exe in Program Files after installation
• Have existing configs/stats/blacklists in Program Files or the local launch directory
• Refuse UAC prompts at any elevation step
• Scheduled tasks missing, renamed, or corrupted
• Multiple copies of the exe in different directories
• Running from unusual locations (network drives, paths with spaces/unicode)

Step 2 — Test Combinatorial Paths

For each pair/group of options, trace the user journey through the code. Use query_callchain.ps1 to follow how config option reads flow through functions across files. Use query_ipc.ps1 to check cross-process message flows that can trigger option interactions. Use query_function_visibility.ps1 <funcName> to find all callers of elevation/UAC functions. Use query_global_ownership.ps1 <globalName> to trace install state ownership across files:

• How does feature X behave when run as admin vs not admin?
• What happens if UAC is refused at each elevation prompt?
• What if the user has option A enabled but then changes option B?
• What if the exe is already running when an install/update/config change happens?
• What state is left behind if a multi-step operation fails halfway?

Focus on state transitions that span multiple steps — wizard → install → create task → relaunch. If any step fails, is the state recoverable? Does the next launch detect the partial state correctly?

By-Design Choices (Do NOT Flag)

These are intentional — do not propose changing them:

• Install path is hardcoded to Program Files
• Exe name is hardcoded when installed to Program Files

Issues around or caused by these choices (e.g., "what if the user renames the installed exe and then auto-update tries to find it") are real findings. The design choices themselves are not.

Step 3 — DRY Consolidation

Multiple bugs in these flows have come from duplicated logic across files. After identifying bugs, also look for:

• Same path-resolution logic implemented differently in multiple places
• Same "is admin?" / "is installed?" / "is running?" check duplicated with slight variations
• Elevation patterns that could share a common helper

DRY findings are secondary to bug findings but help prevent future regressions.

Validation

After explore agents report back, validate every finding yourself. These flows have a lot of defensive code that may look like bugs but is handling edge cases intentionally.

For each candidate:

1. Cite evidence: "I verified by reading file.ahk lines X–Y" with actual code quoted. Trace the full path through multiple files if the journey spans them.
2. Trace the user journey: Describe the exact sequence of user actions that triggers the bug (e.g., "User installs to PF → renames exe → launches renamed exe → auto-update runs → ...").
3. Counter-argument: "What existing code handles this?" — Is there a fallback path that catches this case? Does the code already detect and recover from this state?
4. Observed vs inferred: Did you trace the code path through all branches, or infer the bug from one function without checking its callers?

Plan Format

Section 1 — Interaction bugs (specific failure paths):

| Journey | Steps | Failure Point | File:Lines | Impact | Fix | |---------|-------|--------------|-----------|--------|-----| | Install → rename exe → update check | 1. Install to PF 2. User renames exe 3. Auto-update downloads | Update writes to hardcoded name, old renamed exe orphaned | update.ahk:142 | Stale exe left behind | Check actual exe name, not hardcoded |

Section 2 — UAC refusal gaps (what happens when elevation is denied):

| Operation | UAC Point | Behavior if Refused | Clean? | Fix if Needed | |-----------|----------|-------------------|--------|--------------| | Enable admin mode | Task creation | Falls back to non-admin | Yes | None | | Install to PF | File copy | ??? | ??? | ??? |

Section 3 — DRY consolidation opportunities:

| Pattern | Files | Variations | Proposed Consolidation | |---------|-------|-----------|----------------------| | "Is installed to PF?" check | launcher_main.ahk:30, update.ahk:55, wizard.ahk:80 | Slightly different path comparisons | Shared IsInstalledToProgramFiles() helper |

Order by severity: data-loss risks first, cosmetic inconsistencies last.

Ignore any existing plans — create a fresh one.