Reelyze

intent

reelyze is an ai frame-by-frame analyzer for short-form video creators. given a public instagram reel, tiktok, or youtube short url, it runs a full performance audit (hook strength, retention curve, exact drop-off seconds, scene-by-scene breakdowns, and specific fixes), transcribes the audio, downloads the video as mp4, extracts audio as mp3, or generates viral scripts and content ideas tailored to a creator's niche. use this skill when a user wants to score or audit a video, fix retention issues, get a transcript, download a reel, or generate new content ideas from a topic or creator angle.

inputs

environment variables (required for all operations)

• REELYZE_API_KEY (required): api key from reelyze dashboard, format rk_live_.... set once, send as Authorization: Bearer ${REELYZE_API_KEY} on every request.
• REELYZE_BASE_URL (optional): defaults to https://api.getreelyze.com. override if using a custom endpoint.

external connection

• reelyze api: https://api.getreelyze.com. cors enabled, bearer-token auth. no account linking required (video urls must be public).

user inputs (context-dependent)

• video url: public instagram reel, tiktok, or youtube short link (required for analyze, transcript, download, audio tools).
• topic: freeform text describing the video topic (required for script generation; optional for ideas if using niche context).
• duration (optional for script): video length in seconds. defaults to 30.
• language (optional): defaults to english. supported on transcript, script, ideas.
• count (optional for ideas): number of ideas to generate. defaults to 6.

rate limits and quotas

• free tools (transcript, download, audio): 50 calls per day per api key.
• analyze: requires paid plan. monthly quota by tier: creator 20, pro 60, studio 200, enterprise 400.
• script / ideas: requires paid plan. shared monthly content-generation quota by tier: creator 50, pro 150, studio 600.
• rate limit (global): standard api rate limits apply. 429 if quota exceeded, 402 if plan is free.

procedure

step 1: validate api key and user input

1.1 read REELYZE_API_KEY from environment.

1.2 if not set, ask user to provide it and store it (warn that it will be sent on every request).

1.3 read REELYZE_BASE_URL from environment, default to https://api.getreelyze.com.

1.4 validate the user's input:

• if analyzing, transcribing, downloading, or extracting audio: user must provide a public video url.
• if generating a script: user must provide a topic.
• if generating ideas: optional topic; ideas are richer with niche context.

1.5 output: validated api key, base url, and normalized user parameters (url, topic, language, etc.).

step 2: submit job (for video tools: analyze, transcript, download, audio)

2.1 for analyze: post to ${REELYZE_BASE_URL}/v1/analyze with body {"url":"<user's video url>"}.

2.2 for transcript: post to ${REELYZE_BASE_URL}/v1/transcript with body {"url":"<user's video url>","language":"en"} (language optional, defaults to auto-detect).

2.3 for download (mp4): post to ${REELYZE_BASE_URL}/v1/download with body {"url":"<user's video url>"}.

2.4 for audio (mp3): post to ${REELYZE_BASE_URL}/v1/audio with body {"url":"<user's video url>"}.

2.5 include header Authorization: Bearer ${REELYZE_API_KEY} and Content-Type: application/json.

2.6 handle http errors:

• 401: invalid, missing, or revoked api key. ask user to verify the key.
• 402: user's plan does not include this tool (e.g., free plan trying to analyze). explain the tier required.
• 403: api key lacks permission. regenerate or contact reelyze support.
• 429: quota exhausted or rate limit hit. tell user to wait or upgrade.
• 503: job queue temporarily unavailable. retry after 5 seconds.
• 4xx or 5xx other: log the status and message, tell user to retry or contact support.

2.7 parse response: extract job_id from the json response. response format: {"job_id":"<id>","status":"queued","tool":"<tool>","poll":"/v1/jobs/<id>"}.

2.8 output: job_id, tool name, and status.

step 3: poll job until completion (for video tools only)

3.1 poll GET ${REELYZE_BASE_URL}/v1/jobs/${job_id} every 3 seconds for up to 3 minutes (60 attempts).

3.2 include header Authorization: Bearer ${REELYZE_API_KEY}.

3.3 on each poll, check the status field:

• queued or processing: continue polling. do not fabricate or return partial results.
• completed: proceed to step 4 to extract the result.
• failed: read the error field and report it to the user verbatim. explain that the video may be private, too long (over ~3 minutes), or unreachable. do not retry automatically.

3.4 timeout handling: if 3 minutes pass and status is still processing, tell the user the job is still running, provide the job_id, and suggest they check back later or contact support.

3.5 output: final status, result fields (if completed), or error message (if failed).

step 4: extract and format result (for video tools)

4.1 for analyze: read report_markdown from the completed job response. the report contains: verdict, hook strength, first-3-second assessment, scene-by-scene breakdown with retention signals, drop-off moments with exact seconds, and fixes. summarize for the user (see step 4.5 for formatting rules). do not invent scores or data.

4.2 for transcript: read transcript or transcript_text field. return the plain text to the user.

4.3 for download (mp4): read download_url or artifact_url field. warn the user that the link is temporary (expires after a few hours). provide the link for download.

4.4 for audio (mp3): read artifact_url or download_url field. warn the user that the link is temporary. provide the link for download.

4.5 summarization rules for analyze reports:

• lead with the verdict and the single biggest fix.
• include the exact second(s) where viewers are predicted to drop off.
• quote the report; do not paraphrase or invent data.
• highlight hook strength and first-3-second retention assessment.
• list scene-by-scene retention signals for negative (drop-off) moments.
• if the report is very long, summarize the top 3 fixes and defer to the full report for details.

4.6 output: formatted result text, download link (if applicable), or error message.

step 5: synchronous content generation (script and ideas)

5.1 for script generation: post to ${REELYZE_BASE_URL}/v1/script with body {"topic":"<user's topic>","duration_seconds":<seconds>,"language":"<language>"} (duration and language optional, defaults to 30 and english).

5.2 for ideas generation: post to ${REELYZE_BASE_URL}/v1/ideas with body {"count":<count>,"language":"<language>"} (count defaults to 6, language defaults to english).

5.3 include header Authorization: Bearer ${REELYZE_API_KEY} and Content-Type: application/json.

5.4 handle http errors (same as step 2.6): 401, 402, 403, 429, 503, etc.

5.5 parse response directly (no polling):

• script response: {"script":{"hooks":[...], "script":[{timecode, label, spoken, on_screen},...], "caption":"...", "hashtags":[...], "shot_list":[...]}}
• ideas response: {"ideas":{"ideas":[{title, angle, hook, why_it_works, format},...], "summary":"..."}}

5.6 output: formatted script (hooks, timed script segments, caption, hashtags, shot list) or ideas (list of tailored ideas with hooks and reasoning).

decision points

if user provides a video url:

• if user asks for a full performance audit (keywords: "audit", "analyze", "why did it flop", "retention", "drop-off", "hook strength"): route to analyze (step 2.1, 3, 4.1).
• else if user asks only for the spoken words or transcript (keywords: "transcript", "what did they say", "words"): route to transcript (step 2.2, 3, 4.2). transcript is free and faster.
• else if user asks to download or save the video: route to download (step 2.3, 3, 4.3).
• else if user asks to extract the audio or music: route to audio (step 2.4, 3, 4.4).

if user asks for script generation:

• user must provide a topic (or ask user for one).
• call script endpoint (step 5.1, 5.5). no polling.

if user asks for content ideas:

• user may provide a topic or niche context. if neither is provided, call ideas endpoint with minimal context (ideas are richer with niche data, but still work).
• call ideas endpoint (step 5.2, 5.5). no polling.

if api key is missing:

• ask user to provide it. do not proceed without it.

if api key is invalid (401):

• tell user to verify the key format (rk_live_...) and regenerate if needed.

if plan is free and user requests a paid tool (analyze, script, ideas):

• return 402 error and explain that a paid plan (creator, pro, studio, or enterprise) is required.

if quota is exhausted (429):

• tell user they have hit their monthly limit for this tool. suggest upgrading or waiting until the next month.

if job fails during polling:

• check the error message. common causes: private video, duration over 3 minutes, unreachable url. report the error verbatim; do not retry.

if job does not complete within 3 minutes:

• tell user the job is still processing. provide the job_id and suggest checking back later.

if download or audio link is returned:

• always warn the user that the link is temporary (expires after a few hours) and should be downloaded or shared immediately.

output contract

success criteria:

• analyze: return markdown report summary including verdict, hook assessment, first-3-second retention, exact drop-off seconds, top fixes. quote the report. do not invent scores.
• transcript: return plain text of the spoken words and any on-screen text or captions.
• download (mp4): return a direct, temporary https link to the mp4 file. include expiration warning.
• audio (mp3): return a direct, temporary https link to the mp3 file. include expiration warning.
• script: return a structured script with hooks array, timed script segments (timecode, label, spoken words, on-screen text), caption, hashtags array, and shot list.
• ideas: return a list of content ideas with title, angle, hook, why_it_works explanation, and format.

output format

• all results are text or structured json. markdown reports are returned as markdown. scripts and ideas are returned as json or formatted text.
• all urls (download, audio, artifact) are https and temporary (do not persist or rely on them after a few hours).
• all errors include an http status code, error message, and user-facing guidance.

file locations

• temporary files (mp4, mp3 downloads): stored on reelyze cdn, served via https link. not persisted on user's machine unless they download.
• job metadata: stored on reelyze servers; job_id can be used to check status or retry polling if needed.

outcome signal

the user knows the skill worked when:

• after analyze: the user receives a summarized verdict, the exact second(s) where viewers drop off, the biggest fix to apply, and a recommendation based on the report (e.g., "your hook is weak, rewrite the first 3 seconds").
• after transcript: the user receives the full spoken transcript and any visible on-screen text, word-for-word.
• after download or audio: the user receives a direct https link, downloads the file successfully, and sees a working mp4 or mp3.
• after script generation: the user receives a complete, timed script with multiple hooks, a shot list, captions, and hashtags ready to use.
• after ideas generation: the user receives a list of 6 (or requested count) tailored content ideas, each with a hook, angle, and reasoning for why it works.
• error states: if something fails, the user sees a clear error message (e.g., "video is private", "plan is free, requires creator tier or higher", "quota exhausted, retry next month") and knows what to do next (verify url, upgrade plan, etc.).
• polling feedback: if a job takes longer than expected, the user is told the job is still running, not hung, and given a job_id for reference.