refactor: extract shared tool boilerplate in tools.ts (plan 018)

[DavertMik]

Plan 018 — Consolidate duplicated boilerplate in tools.ts

src/ai/tools.ts (the Tester's tool belt) is built from copy-paste. This extracts the shared shapes so each tool contains only what's tool-specific. Behavior-preserving.

Done

• errorText(error) — replaces the 9 identical error instanceof Error ? error.toString() : 'Unknown error occurred' ternaries (banned by CLAUDE.md).
• commitNote(activeNote, result, toolResult, action) — replaces the 9 screenshot-guard + commit pairs; a fix to one copy previously missed eight. (The form PASSED path's screenshot-then-branch was proven equivalent: !hasObservablePageChange ⟹ the aria/url guard is false.)
• 3 conditional spreads ...(action.lastError && { error }) → plain if.
• createIframeTools(explorer) — renamed from createSpecialContextTools(explorer, 'iframe'); deletes the unreachable context !== 'iframe' guard. Tester call site's banned conditional spread → plain if (isInsideIframe) Object.assign(...); exitIframe is unique so assigning last is equivalent.
• createLearnExperienceTool({ experienceTracker, getState }) factory — removes Navigator's null as unknown as Researcher type-lie.
• getVisitedStates routes through successToolResult, gaining the action field.
• First unit tests for tools.ts (TEST-01) — validation paths + factories, no browser.

Deferred / coordinated

• Step 4 (click/hover fallback-loop) — DEFERRED: click/hover are not pure clones (click screenshots on success + disambiguates; hover does neither); folding risks subtle change without a live browser.
• Step 8 (disambiguateElements → Navigator) — DEFERRED (optional, high churn).
• Step 6 (countAriaChanges → utils/aria.ts) — handled by plan-015 PR refactor: simplify ActionResult loading and Diff API (plan 015) #87; not duplicated here.
• currentActionResult alias — deferred (cosmetic; createAgentTools declares stateManager per-execute).

Verification

• bun test tests/unit → 770 pass / 0 fail (+6 new).
• bun test tests/integration/ → 63 pass / 0 fail.
• bun run lint → clean; tsc in touched files 4 → 3, no new errors.

🤖 Generated with Claude Code

[github-actions]

Explorbot Self-Regression

Commit 2a697a5 · run

Scenario	Result	Attempts	Duration
basic (native)	PASS	1/3	6m
experience: control	OK — failed as expected	1/1	1m
experience: seeded	PASS	1/3	51s

Attempt details

• basic (native) attempt 1 — PASS: login evidence: PASS (post-login plan=true, post-login research=true); research: PASS (files=5, wellFormed=true, keywords=6/3); scenarios: PASS (tests=7/5, features=4/3); tests passed: PASS (5 passed, 0 failed (reporter: 5 passed, 0 failed))
• experience: control attempt 1 — PASS: control: OK — failed as expected (0 passed, 1 failed)
• experience: seeded attempt 1 — PASS: seeded: PASS (1 passed, 0 failed)

---

Session analysis — basic (native):

Session Analysis

The issue tracking feature was tested for core operations: creation, filtering by label, keyword search, status filtering, and detail view. The keyword search and status filtering work correctly. The label filter test has a verification gap—it confirms the URL changes but does not verify the actual filtered results.

Coverage

• Pages: /issues, /issues/{id}
• Features: Issue creation, label filtering, keyword search, status filtering, issue detail view

What works

• Issue creation — ET-1 Create a new issue with title and verify it appears in the list
• Keyword search — ET-3 Search for an issue by keyword and verify filtered results
• Status filtering — ET-4 Filter issues by status and verify only matching issues show
• Issue detail view — ET-5 Open an existing issue to view its details

Execution Issues

• ET-2 Filter issues by 'bug' label and verify only bug issues are shown — passed vacuously, only verified URL contains ?label=1, did not verify the filtered list actually shows only bug-labeled issues