claude-appeal-evidence-writer

name

claude-appeal-evidence-writer

description

Generate privacy-safe, evidence-backed appeal letters for Claude or Anthropic account suspensions using local Claude data exports. Use this skill whenever the user wants to appeal a Claude account ban, automatically find local Claude export data, summarize Claude usage history for support, use local Claude export data as evidence, draft a Trust & Safety escalation letter, prepare Google Form appeal text, or request access/export of user-generated Claude content. Do not use this skill to automate form submissions, spam appeals, bypass reCAPTCHA, evade bans, rotate identities, hide account linkage, or misrepresent account activity.

Claude Appeal Evidence Writer

Use this skill to help a user prepare a legitimate account appeal using local Claude export data. Keep the work framed as fair review, account reinstatement, and access to user-generated content. Do not help with ban evasion or automated complaint spam.

Workflow

1. Start by trying the automated preparation script. It scans only common local locations: the current directory, parent directory, Downloads, Desktop, and Documents.

python3 scripts/prepare_appeal.py --out outputs

2. If the script finds one or more Claude exports, use the selected export in its JSON output. The script chooses the most recently modified candidate by default and lists all candidates. If it finds no export, ask the user for a local Claude export zip or extracted folder path. If multiple candidates look surprising, list them and confirm before continuing.
3. Review the generated evidence files, especially outputs/evidence/evidence-summary.md, for:
   • latest substantive conversation,
   • recent normal-use topics,
   • usage categories,
   • conversation/project counts,
   • privacy redaction warnings.
4. Ask only for missing appeal fields surfaced in next_questions:
   • account email,
   • contact email, defaulting to account email when appropriate,
   • name/signature,
   • approximate suspension date, if known,
   • desired tone, defaulting to firm,
   • whether to include a Chinese-language or identity risk-profiling concern.
5. Once the user gives those details, run the same preparation script with the selected export and filled fields:

python3 scripts/prepare_appeal.py \
  --out outputs \
  --export-path /path/to/selected-claude-export.zip \
  --account-email [e-mail] \
  --contact-email [e-mail] \
  --name "User Name" \
  --suspension-date 2026-07-01 \
  --tone firm \
  --include-language-concern

6. Give the user paths to:
   • outputs/appeal/appeal-letter.md for email or support escalation,
   • outputs/appeal/google-form-message.txt for manual form submission.
7. Remind the user to manually review and submit. Never click Submit, bypass reCAPTCHA, or run unattended form submission.

Safety Boundaries

• Process export files locally. Do not upload Claude exports or generated evidence to third-party services.
• The automated search is intentionally narrow. Do not scan the whole home directory unless the user explicitly asks and understands the privacy/performance tradeoff.
• Report summaries, counts, categories, dates, and short redacted snippets. Do not expose full conversations by default.
• Never print token values, cookies, stable IDs, phone numbers, or full emails from users.json.
• Do not generate false claims. Use the export to ground concrete normal-use examples.
• If the user asks for automatic daily submissions, scraping, identity rotation, or ban evasion, refuse that part and offer a manual reminder or one-click draft workflow instead.

Manual Fallback

If automated preparation is not available in the agent environment, run the lower-level scripts directly:

python3 scripts/extract_claude_export.py /path/to/claude-export.zip --out outputs/evidence
python3 scripts/compose_appeal.py outputs/evidence/appeal-facts.json --out outputs/appeal --account-email [e-mail]

Writing Guidance

Read references/appeal-writing.md before drafting or revising appeal language by hand. Use the scripts for deterministic extraction and first drafts, then edit the final letter for the user's exact situation.