feat(export): Phase 7 - Export and Lab Trust Pass

[d-oit]

Description

Modernize Export panel with preview-first UX and clarify AI Harness status.

Tasks

ExportPanel.tsx

• Replace generic panel copy with preview-first export UX
• Add format cards: Markdown, JSON, SQLite snapshot, Graph data, Static Site
• Add scope selector: all, selected entity, graph neighborhood, search results
• Add counts before export (e.g., "12 entities, 47 claims, 156 links")
• Add progress and success/failure states
• Explain that SQLite/OPFS is canonical and Markdown/JSON/site are exports

AIHarness.tsx

• Move under Lab or mark experimental
• Stop reusing .chat-view if it becomes an agent console
• Add truthful state: local model unavailable, local-only, or provider configured
• Disable non-functional send action until behavior exists
• Use panels: Prompt, Context, Run Log, Artifacts

JobMetrics.tsx

• Move inline styles into tokenized CSS
• Gate as debug-only if not a product surface
• Consider product job queue indicator: indexing, exporting, stale index, failed jobs

Acceptance Criteria

• Export does not look complete until artifacts are available
• AI Harness does not imply working assistant when none exists
• Local-first boundaries visible and truthful
• Copy does not mention DB, console, or "agent ready"

Anti-Slop Rules

• No vague AI badges
• No implementation words in product UI
• No toasts as only feedback for saving/indexing/export

References

• plans/UI_MODERNIZATION_2026.md Phase 7

[github-actions]

👋 This looks like a duplicate of #78.

The canonical issue has broader scope and/or more tasks — please subscribe there for updates.

⏳ This issue will be auto-closed in 3 days unless you react with 👎 to dispute.

Detected via title similarity analysis

[d-oit]

@jules address issue in a new pr.

You are an autonomous PR repair agent. Work through every phase in order. Never ask for confirmation — act, then report.

---

## PHASE 0 — BOOTSTRAP

1. Load agent instructions:
   - Read `agents.md` in the repo root. Follow all instructions there exactly and for the remainder of this session.
   - If `agents.md` is not found, check `.claude/agents.md`. If neither exists, note it and continue.

2. Load existing learnings:
   - Check for `plans/learnings.md` or any `*learnings*` file under `plans/`.
   - Read and internalize all entries before proceeding.

3. Detect agent skills:
   - Scan `/mnt/skills/` (public, private, user, examples) for any SKILL.md relevant to this repo's language/stack.
   - Load and apply matching skills on demand throughout this session.

4. Ensure `plans/` directory exists:
   ```bash
   mkdir -p plans

---

PHASE 1 — ENVIRONMENT AUDIT

git status
git log --oneline -10
git branch -a
git remote -v
gh pr view --json number,title,body,reviews,comments,reviewRequests,mergeable,mergeStateStatus
gh pr checks

Capture: current branch, upstream branch, PR number, merge state, CI job status.

---

PHASE 2 — MERGE CONFLICT RESOLUTION

1. Identify all conflicts:

   git diff --name-only --diff-filter=U

2. For each conflicted file:
   • Read the full conflict region (<<<<<<< … ======= … >>>>>>>).
   • Resolve by applying both sides' intent unless mutually exclusive — in that case prefer the PR branch change and document the decision in plans/issues.md.
   • Never leave conflict markers in the final file.
3. After all conflicts are resolved:

   git add .
   git merge --continue   # or rebase --continue if in rebase flow

---

PHASE 3 — PR FEEDBACK ANALYSIS & FIXES

1. Fetch all review comments (inline + general):

   gh pr view --json reviews,comments
   gh api repos/{owner}/{repo}/pulls/{pr_number}/comments

2. For each comment/suggestion:
   • Categorize: [CODE] [STYLE] [TEST] [DOCS] [CONFIG] [SECURITY] [PERF]
   • Apply the fix directly in the affected source file.
   • If a suggestion is ambiguous, implement the most defensive/correct interpretation.
   • If a fix is not possible (external dependency, out of scope, requires human decision), append to plans/issues.md:

     ## NON-FIXABLE: <short title>
     File: <path>
     Reviewer comment: "<exact quote>"
     Reason not fixed: <explanation>
     Action required: <what a human must do>

---

PHASE 4 — CI JOB FAILURE ANALYSIS & FIXES

Important: Fix the source code, tests, and project configuration that cause CI jobs to fail.
Do NOT modify .github/workflows/ files or any CI pipeline definitions unless a workflow file itself has a syntax error introduced by this PR's changes.

1. Read CI config to understand what each job does — for reference only:

   cat .github/workflows/*.yml 2>/dev/null

2. Fetch the current CI job failures:

   gh pr checks
   gh run list --branch $(git branch --show-current) --limit 5
   gh run view <run-id> --log-failed

3. For each failing job, identify the root cause in source files and fix it:

   Job type	Fix target
   Lint / format	Fix source files using the project's linter/formatter (eslint --fix, cargo fmt, black, gofmt, prettier --write, etc.)
   Typecheck	Fix type errors in source files; do not use any, @ts-ignore, or equivalent blanket suppression
   Unit / integration tests	Fix the code under test, not the test itself, unless the test is provably incorrect
   Build / compile	Resolve import, dependency, or configuration issues in source files
   Security scan	Remove secrets, fix vulnerable dependency versions in package.json/Cargo.toml/requirements.txt etc.
   Coverage threshold	Add missing tests to cover newly introduced code paths

4. After fixing each category, re-run the relevant check locally to confirm:

   # adapt to stack — examples:
   cargo test && cargo clippy
   npm test && npx tsc --noEmit && npx eslint .
   python -m pytest && python -m pylint src/

5. If a CI job failure cannot be fixed without human intervention, append to plans/issues.md:

   ## NON-FIXABLE CI FAILURE: <job name>
   Run: <gh run URL>
   Error: <exact error message>
   Reason not fixed: <explanation>
   Action required: <what a human must do>

---

PHASE 5 — WARNINGS & PRE-EXISTING ISSUES

1. Run a full warning scan appropriate to the project stack:

   cargo clippy 2>&1 | grep -E "^warning"
   npx tsc --noEmit 2>&1
   npx eslint . 2>&1
   python -m pylint src/ 2>&1

2. Fix all warnings that are safe to fix without changing behavior.
3. For warnings that cannot be safely fixed, append to plans/issues.md.
4. Scan for common pre-existing issues:
   • Unused imports / dead code → remove
   • Hardcoded secrets or credentials → replace with env var references and document in plans/issues.md
   • Missing error handling on I/O operations → add minimal safe handling
   • TODO/FIXME comments predating this PR → note in plans/issues.md, do not delete

---

PHASE 6 — COMMIT & PUSH

Commit in logical atomic groups (one commit per concern):

git add <specific files>
git commit -m "<type>(<scope>): <description>

- <bullet of what changed>
- <bullet of why>

Fixes: <reviewer comment summary or CI job name>"

Commit types: fix feat refactor docs test ci chore style

After all commits:

git push origin HEAD

If push is rejected due to remote changes:

git pull --rebase origin <upstream-branch>
# resolve any new conflicts → repeat Phase 2
git push origin HEAD

---

PHASE 7 — POST-PUSH VERIFICATION

gh pr checks --watch   # wait for all CI jobs to complete
gh pr view --json mergeable,mergeStateStatus,reviews

• If new CI job failures appear after push, return to Phase 4 and fix them.
• Confirm all review threads are addressed.

---

PHASE 8 — LEARNINGS UPDATE

1. Open or create plans/learnings.md.
2. Append a compact dated entry:

   ## <YYYY-MM-DD> — PR #<number>: <title>
   
   ### What was fixed
   - <item>
   
   ### CI jobs fixed
   - <job name>: <root cause and fix summary>
   
   ### Patterns to remember
   - <reusable insight>
   
   ### Non-fixable issues documented
   - <count> items in plans/issues.md
   
   ### Skills used
   - <skill name>: <why it was applied>

3. Compact any existing entries older than 30 days into a single ## Archive section — preserve key patterns, remove verbose details.

---

CONSTRAINTS

• Follow agents.md instructions above all else (except safety).
• All plans, issues, and learnings files are written to plans/ only.
• Never modify .github/workflows/ files to make CI pass — fix the source instead.
• Never force-push to main or master.
• Never delete tests to make them pass.
• Never suppress type errors with blanket ignores without documenting in plans/issues.md.
• Never modify plans/issues.md entries once written — only append.
• All file edits must be syntactically valid — run a parse/compile check after each edit before committing.
• If a skill is loaded, follow its instructions exactly.
• Keep each commit under 400 lines changed where possible; split larger changes.

---

OUTPUT FORMAT (report after completion)

## PR Agent Report

**PR:** #<n> — <title>
**Branch:** <name>
**Commits pushed:** <n>

### ✅ Fixed
| # | Category | File | Description |
|---|----------|------|-------------|

### 🔧 CI Jobs
| Job | Before | After | Fix applied |
|-----|--------|-------|-------------|

### ⚠️ Non-fixable (see plans/issues.md)
| # | File | Reason |
|---|------|--------|

### 📚 Learnings updated
- plans/learnings.md — <n> entries added, <n> compacted

[google-labs-jules]

Jules is on it. When finished, you will see another comment and be able to review a PR.

[google-labs-jules]

Ready for a review! A PR has been created.