audible-goodreads-deal-scout（有声好读交易侦察）

Audible Goodreads Deal Scout

Never miss an Audible deal for a book you actually want to read.

Audible Goodreads Deal Scout is a ClawHub / OpenClaw skill that checks Audible promotions against Goodreads ratings, your Goodreads shelves, and optional reading-taste notes. It is built for people who want fewer random deal notifications and more "this is actually relevant to me" recommendations.

The goal is not to buy more audiobooks. The goal is to notice the few Audible deals that match books you already care about.

What It Does

Use this skill to:

• check today's Audible daily deal and get a short recommendation
• see whether the deal is already on your Goodreads Want-to-Read shelf
• suppress books you already marked as read or currently reading
• scan your Goodreads Want-to-Read shelf for Audible titles with visible discounts
• rank discounted Want-to-Read matches by deal strength and Goodreads signal
• create personalized recommendations from your Goodreads shelves, ratings, reviews, and taste notes
• explain why a book may or may not fit your reading preferences, instead of only showing a generic score
• optionally check member-visible Audible cash prices with a local Audible auth token
• optionally deliver good matches to Telegram or another configured OpenClaw channel

If you do not want personalization, the skill still works with public Goodreads ratings only.

This skill reports opportunities only. It does not buy, reserve, check out, redeem credits, manage subscriptions, or complete purchases.

Getting The Most Value

The best setup uses:

1. A Goodreads library CSV from the official Goodreads Import and Export page.
2. Short taste notes about books, authors, genres, pacing, and themes you like or avoid.
3. Optional Audible authentication when you want member-visible cash prices in Want-to-Read scans.

With those pieces, the skill can track today's Audible promotion, compare it against your Goodreads library, explain personal fit, and scan your Want-to-Read shelf for Audible titles that look discounted or priced below list.

Trust And Data Access

The skill reads only the files and services needed for its configured workflow.

The local auth file is sensitive. Do not paste its contents into chat, commit it, or publish it.

For a fuller trust and data-access summary, see TRUST.md in the published bundle or repository.

Optional Audible Authentication

Audible authentication is optional and separate from the normal daily-deal workflow. The skill works without it, but anonymous Audible pages often hide cash prices behind credit or membership UI.

Use authenticated price lookup only when the user explicitly wants member-visible cash prices in Want-to-Read scans. Do not ask for the user's Audible or Amazon password. Use audible-auth-start, audible-auth-finish, and audible-auth-status instead.

Agent Runtime Instructions

Use this skill when the user wants to:

• set up a reusable Audible deal workflow
• check the current Audible daily promotion
• scan Goodreads Want-to-Read books for visible Audible US discounts
• personalize the result with a Goodreads CSV and/or freeform notes
• finalize and optionally deliver the result into a configured channel

Runtime output contract

The skill runtime must return JSON only in this shape:

{
  "schemaVersion": 1,
  "goodreads": {
    "status": "resolved | no_match | lookup_failed",
    "url": "string | null",
    "title": "string | null",
    "author": "string | null",
    "averageRating": "number | null",
    "ratingsCount": "integer | null",
    "evidence": "string | null"
  },
  "fit": {
    "status": "written | not_applicable | unavailable",
    "sentence": "string | null"
  }
}

Use the Python prep layer first

Do not fetch Audible yourself in model text. Always start with the prep layer:

sh "{baseDir}/scripts/audible-goodreads-deal-scout.sh" prepare --config-path "<config-path>" --invocation-mode manual

Prep returns JSON with:

• status: ready | suppress | error
• reasonCode
• warnings[]
• audible
• personalData
• artifacts
• metadata

If prep returns suppress or error, surface that result directly and stop. Do not do Goodreads lookup or fit writing after a prep-layer short-circuit.

Setup

If the skill is not configured yet, gather:

1. Audible store
2. Whether the user wants personalization
3. Optional Goodreads CSV path
4. Optional notes file or pasted notes
5. Optional threshold override from the default 3.8
6. Optional delivery channel and target
7. Delivery policy: positive_only, always_full, or summary_on_non_match
8. Whether daily automation should be enabled

Then write config through:

• If the user does not request a custom path, use the default workspace-root storage path: <workspace>/.audible-goodreads-deal-scout/.
• Do not invent legacy names like .audible-goodreads-deal.
• Do not store mutable config, state, or artifacts inside {baseDir} or the installed skill folder. openclaw skills install and openclaw skills update --force replace the workspace skill directory.

sh "{baseDir}/scripts/audible-goodreads-deal-scout.sh" setup \
  --config-path "<config-path>" \
  --audible-marketplace "<marketplace>" \
  --threshold "<threshold>" \
  [--goodreads-csv "<csv-path>"] \
  [--notes-file "<notes-file>"] \
  [--notes-text "<inline notes>"] \
  [--delivery-channel "telegram"] \
  [--delivery-target "<target>"] \
  [--delivery-policy "positive_only"] \
  [--daily-automation] \
  [--register-cron]

Use interactive setup only when the user explicitly wants prompt-by-prompt CLI onboarding. Otherwise prefer the non-interactive command with concrete flags.

Want-to-Read discount scan

Use this only when the user asks to scan Goodreads Want-to-Read books for Audible discounts. This is a manual audit command, not a cron or delivery workflow.

Requirements:

• The configured Goodreads CSV must exist.
• V1 supports Audible US only.
• Do not create cron jobs or send delivery messages for this command.
• Do not perform extra model web searches; the Python command handles Audible lookup and report generation.
• If setup or runtime state is unclear, run doctor before retrying.

Run:

sh "{baseDir}/scripts/audible-goodreads-deal-scout.sh" scan-want-to-read \
  --config-path "<config-path>" \
  [--audible-auth-path "<auth-path>"] \
  [--limit 40] \
  [--offset 0] \
  [--scan-order newest] \
  [--progress plain] \
  [--no-goodreads-rating-enrichment] \
  [--goodreads-rating-limit 20] \
  [--output-json "<json-path>"] \
  [--output-md "<markdown-path>"]

Default behavior:

• Print compact Markdown to stdout.
• Write progress to stderr by default. Use --progress json for machine-readable JSONL progress or --progress none when silence is required.
• Show visible numeric discounts first.
• Suppress long non-deal lists unless --include-non-deals is requested.
• Suppress duplicate Audible product matches in the final report while preserving scanned-row counts.
• Use --offset and --limit for large Goodreads backlogs.
• For long agent-run scans, prefer --progress json plus --output-json and --output-md so progress logs and final reports stay separate.
• Use pricing.priceBasis and pricing.dealType to distinguish member cash prices below list from true limited-time sale or promotion signals.
• If CSV average ratings are missing, the command may enrich a small number of discounted rows from public Goodreads book pages by Goodreads book id. Disable with --no-goodreads-rating-enrichment when the user wants no Goodreads page fetches.

Important caveat: Audible often hides cash prices behind credit or membership UI. Treat price_hidden, price_unknown, and needs_review as honest uncertainty, not as failures.

Optional authenticated pricing:

• If the user asks for headless authenticated Audible prices, use audible-auth-start and audible-auth-finish.
• Do not ask for the user's Audible or Amazon password.
• Store the auth file under the workspace storage directory, for example <workspace>/.audible-goodreads-deal-scout/audible-auth.json.
• Treat the auth file as sensitive and never paste its token contents into chat.
• Authenticated scans usually spend one search request plus one authenticated price request for each matched title; set --max-requests accordingly.
• Treat cash pricing fields as the source of truth and do not classify Audible credit prices, including credit_price, as cash discounts.
• Treat authenticated discounted as "member-visible cash price below list price", not proof of a limited-time sale; check pricing.dealType.
• Use audible-auth-status to check readiness, expiry, and file permissions without exposing tokens.

sh "{baseDir}/scripts/audible-goodreads-deal-scout.sh" audible-auth-start \
  --auth-path "<workspace>/.audible-goodreads-deal-scout/audible-auth.json" \
  --audible-marketplace us

sh "{baseDir}/scripts/audible-goodreads-deal-scout.sh" audible-auth-finish \
  --auth-path "<workspace>/.audible-goodreads-deal-scout/audible-auth.json" \
  --redirect-url "<final-amazon-redirect-url>"

Troubleshooting:

sh "{baseDir}/scripts/audible-goodreads-deal-scout.sh" doctor \
  --config-path "<config-path>"

sh "{baseDir}/scripts/audible-goodreads-deal-scout.sh" audible-auth-status \
  --auth-path "<workspace>/.audible-goodreads-deal-scout/audible-auth.json"

Ready flow

For ready_* prep results:

1. Read artifacts.runtimePromptPath
2. Read artifacts.runtimeInputPath
3. Resolve the Goodreads public book page and score with OpenClaw web/search
4. Produce JSON only that matches artifacts.runtimeOutputSchemaPath
5. Finalize through:

sh "{baseDir}/scripts/audible-goodreads-deal-scout.sh" finalize \
  --prepare-json "<prepare-result-path>" \
  --runtime-output "<runtime-output-path>"

If the user wants the result routed to a configured channel:

sh "{baseDir}/scripts/audible-goodreads-deal-scout.sh" run-and-deliver \
  --config-path "<config-path>" \
  --prepare-json "<prepare-result-path>" \
  --runtime-output "<runtime-output-path>"

Decision rules

• If personalData.exactShelfMatch == "to-read", recommend immediately. This overrides the Goodreads threshold.
• If prep already marked the book as read or currently-reading, do not continue.
• Otherwise enforce the Goodreads threshold from metadata.threshold.
• If Goodreads lookup fails, use error_goodreads_lookup_failed.
• If Goodreads cannot confirm a matching book page, use suppress_no_goodreads_match.

Skill-layer reason codes:

• recommend_to_read_override
• recommend_public_threshold
• suppress_below_goodreads_threshold
• suppress_no_goodreads_match
• error_goodreads_lookup_failed

Fit writing

The model writes the fit paragraph. Python does not call a provider API directly.

Use:

• Goodreads public score
• artifacts.fitContextPath
• artifacts.reviewSourcePath when present
• artifacts.notesPath when present

Rules:

• Do not drop rated/reviewed CSV rows for prompt convenience.
• Summarize each written Goodreads review to 500 characters or fewer before using it as evidence.
• If personalData.privacyMode == "minimal", do not use personal CSV or notes content in the fit paragraph.
• If no meaningful personal data exists, say so explicitly instead of inventing taste evidence.

Fallback lines:

• Fit: No personal preference data was configured, so this recommendation is based only on the public Goodreads score.
• Fit: Personalized fit feedback is unavailable right now, but the recommendation decision still completed.

Delivery

run-and-deliver must respect the configured deliveryPolicy:

• positive_only: deliver only recommend
• always_full: deliver the full card for every final status
• summary_on_non_match: deliver full recommend, but a short summary card for suppress or error

For scheduled runs, prep with --invocation-mode scheduled. If prep returns suppress_duplicate_scheduled_run, stop quietly. After a surfaced scheduled result, mark the deal as emitted with:

sh "{baseDir}/scripts/audible-goodreads-deal-scout.sh" mark-emitted \
  --state-file "<state-file>" \
  --prepare-json "<prepare-result-path>"

Use the same current scheduled prepare artifact that was delivered. run-and-deliver refuses scheduled error prep results and stale scheduled artifacts whose metadata.storeLocalDate is not the current Audible marketplace date.