Skill9 starsChanged 7 months agoMITView on GitHub
Fetch everything retrievable from a public Instagram post link — author, engagement, verbatim caption, and the full content of every carousel slide (read visually from downloaded images) — using an anonymous extraction ladder that needs no Instagram login. Use when the user shares an instagram.com/p/ or /reel/ URL and wants its information, content, or slides captured. Do NOT use for X/Twitter links (use the twitter-x skill) or YouTube links (use youtube-transcript).
17 other projects use a version of this.
Adopted by
Most recently taken up by
- proyecto26/instagram-ai-plugintheir own version, first seen Nov 15, 2026
- matthewdonsemail-lab/logtheir own version, first seen Nov 15, 2026
- eumateusalex/Template-OStheir own version, first seen Nov 8, 2026
- AceDataCloud/Skillstheir own version, first seen Nov 8, 2026
- brightdata/skillstheir own version, first seen Nov 1, 2026
- manifoldmcp/marketing-skillstheir own version, first seen Oct 17, 2026
Used with
Projects with it also use tiktok and spotify.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.Other versions17 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 17
- brightdata/skills★ 270same name
- refly-ai/refly-skills★ 204same name
- hyperfx-ai/marketing-skills★ 112same name
- matthewdonsemail-lab/log★ 63same name
- obra/muse-skills★ 20same name
- AceDataCloud/Skills★ 17same name
- moboutrig/instagram-claude-skill★ 10same name
- mediause/agent-skills★ 9same name
- geegewu/gegeewu-open-skill★ 8same name
- jacobwell/muse-skills★ 8same name
7 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository44 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- anthropic-multi-agent-research-system-skillSkill
- apply-selectionsSkill
- bootstrap-dogfood-projectSkill
- brainstormSkill
- branch-choiceSkill
- cc-adoption-scoutSkill
- claude-guardianSkill
- continueSkill
- end-sessionSkill
- executing-plansSkill
- five-advisorsSkill
- get-work-doneSkill
- git-branch-lifecycleSkill
- githubSkill
- grill-meSkill
- hook-transcript-fixture-testSkill
- linkedinSkill
- modern-data-pipeline-diagramSkill
- pattern-qualitySkill
- platform-event-live-probeSkill
- plugin-lifecycleSkill
- promote-to-coreSkill
- prompt-auto-enhanceSkill
- provision-reportSkill
- redditSkill
- review-new-claude-featuresSkill
- save-sessionSkill
- scan-discovery-reportSkill
- scan-repoSkill
- scan-urlSkill
- self-improveSkill
- skill-evaluatorSkill
- skill-masterSkill
- ssot-workflow-auditSkill
- start-sessionSkill
- synthesize-hubSkill
- synthesize-projectSkill
- twitter-xSkill
- verify-effect-at-destinationSkill
- workflow-doc-reviewerSkill
- writing-plansSkill
- writing-skillsSkill
- youtube-transcriptSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
17 rules. 0 are common, 14 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 13 edited versions we read keep these.
14
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 41
No credentials, no paid endpoints, no user decisions needed mid-run.
- Distinctive: line 56
If the URL is not an Instagram post/reel link → HARD-STOP and say which skill fits instead (twitter-x for X, youtube-transcript for …
- Distinctive: line 61
Run WebFetch on the cleaned post URL asking for: author, verified status, date, likes, comments, caption, image alt text.
- Distinctive: line 67
If it returns a login wall: note it in LEARNINGS (rung demotion candidate) and continue — later rungs don't depend on this one.
- Distinctive: line 70
If the user only wanted "what is this post about", STOP here and report.
- Very specific: line 117
Write the collected URLs to <scratchpad>/ig_slide_urls.txt (one per line, exact — the query params are signed and required).
- Distinctive: line 151
After EVERY run, compare what happened against the ladder:
- Very specific: line 185
Always read LEARNINGS.md before running the ladder — Why: the ladder self-corrects there first.
- Distinctive: line 187
Always capture the caption VERBATIM from the embed snapshot before paraphrasing anything — Why: og-meta summaries have been mistaken for …
- Distinctive: line 190
Always verify downloaded slide file sizes before reading — Why: expired signed URLs produce tiny error bodies that waste visual-read calls
- Distinctive: line 192
Always include the Honest read section in the report — Why: the owner's standing rule requires assessment, not just extraction
- Distinctive: line 194
Always run STEP 5 (learning capture) — Why: this is the skill's self-improvement contract.
- Distinctive: line 199
MUST NOT log in to Instagram or ask for credentials — the ladder is anonymous by design.
- Distinctive: line 201
MUST NOT retry a "Known dead ends" path more than once — go to the next rung instead
- Distinctive: line 203
MUST NOT trust alt text for slide content — read the downloaded image
- Very specific: line 204
MUST NOT edit SKILL.md on a single occurrence — record in LEARNINGS.md first.
- Distinctive: line 207
MUST NOT present engagement numbers without their fetch date
“Common” is measured: kept by at least 6 in 10 of the 13 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Instagram — Anonymous Post-Extraction Ladder
Extract a public Instagram post's full information without logging in. Codified from a real successful run (2026-08-06, 8-slide carousel captured end-to-end after 3 dead ends).
Request: $ARGUMENTS
Prerequisites
| Item | Class | Check |
curl | Tool | curl --version |
Playwright MCP (mcp__playwright__*) | Service (rung 2+) | ToolSearch loads the tools |
| Scratchpad directory | Path | listed in system prompt |
| Instagram login | NOT needed | ladder is anonymous by design |
No credentials, no paid endpoints, no user decisions needed mid-run.
STEP 0: Preflight + Read Learnings
- Validate the URL: must match
instagram.com/p/<shortcode>orinstagram.com/reel/<shortcode>. Strip query params (?img_index=,?igsh=) but NOTEimg_index=N— the user linked slide N; call it out in the report. - Read
LEARNINGS.mdin this skill's directory. Apply any rung promotions/demotions and new dead-ends recorded there — the ladder below is the baseline; LEARNINGS.md is the living correction layer. - Probe
curl --version(side-effect-free).
If the URL is not an Instagram post/reel link → HARD-STOP and say which skill fits instead (twitter-x for X, youtube-transcript for YouTube).
STEP 1: Rung 1 — WebFetch the post URL (metadata + caption summary)
Run WebFetch on the cleaned post URL asking for: author, verified status, date, likes, comments, caption, image alt text.
- What this reliably gives: author handle, approximate date, like/comment counts, a SUMMARY of the caption (og-meta description).
- What it does NOT give: verbatim caption, carousel slides beyond #1.
- If it returns a login wall: note it in LEARNINGS (rung demotion candidate) and continue — later rungs don't depend on this one.
If the user only wanted "what is this post about", STOP here and report. Otherwise continue — "all information" means the full ladder.
STEP 2: Rung 2 — Playwright on the embed page (verbatim caption + slide URLs)
The embed page https://www.instagram.com/p/<shortcode>/embed/captioned/
renders anonymously with the FULL caption and a navigable carousel — but
only when real JavaScript runs.
- Load Playwright tools via ToolSearch (
browser_navigate,browser_snapshot,browser_evaluate,browser_close). browser_navigateto the embed URL, thenbrowser_snapshot— the snapshot contains the verbatim caption, author, follower count, likes, and comment count. Capture them.- Carousel: slides lazy-load. Run ONE
browser_evaluatewith an async function that (a) collectsdocument.imageswithnaturalWidth >= 500, (b) clicks the "Next" button, waits ~1200ms, re-collects, repeated up to 10×, (c) returns the deduped{src, alt, w, h}list in insertion order — insertion order == carousel order (verified: a slide stamped "06/08" landed at position 6).
async () => {
const sleep = ms => new Promise(r => setTimeout(r, ms));
const urls = new Map();
const collect = () => { for (const i of Array.from(document.images)) {
const s = i.currentSrc || i.src;
if (i.naturalWidth >= 500 && !urls.has(s)) urls.set(s, {alt: i.alt}); } };
collect();
for (let k = 0; k < 10; k++) {
const b = Array.from(document.querySelectorAll('button'))
.filter(x => ((x.getAttribute('aria-label')||x.textContent||'')).match(/next/i));
if (!b.length) break;
b[0].click(); await sleep(1200); collect();
}
return Array.from(urls.entries()).map(([src, m]) => ({src, ...m}));
}browser_closewhen done.
Single images (no "Next" button) are already complete after step 2.
STEP 3: Rung 3 — Download slides and read them visually
Slide graphics carry the actual content; alt text is auto-generated junk.
- Write the collected URLs to
<scratchpad>/ig_slide_urls.txt(one per line, exact — the query params are signed and required). - Download:
while IFS= read -r u; do curl -s -A "Mozilla/5.0" "$u" -o "slide_$n.jpg"; ...into the scratchpad. Verify non-trivial file sizes (>50KB each; a tiny file = expired/blocked URL → re-collect in Rung 2). Readeachslide_N.jpg(the Read tool renders images) and extract the text/diagram content of every slide.
STEP 4: Compile the report
Locked output format:
## Post details
| Field | Value | <!-- author, verified, followers, date, type, likes, comments, hashtags -->
## Full caption (verbatim)
> <caption exactly as captured in STEP 2>
## All N slides
1. **<slide title>**: <content extracted visually>
...
**Honest read:** <what the post actually is — lead magnet? real content?
inconsistencies between caption promises and slide content>
<risk line: engagement numbers are as-of fetch time>The Honest read section is mandatory — extraction without assessment is half the job (owner standing rule: no sugar-coating).
STEP 5: Capture Learnings (self-learning loop — MANDATORY)
After EVERY run, compare what happened against the ladder:
| Event this run | Action |
| A rung failed that usually works | Append a dated lesson: rung, URL type, exact error |
| A dead end was hit (new blocker) | Append it under "Known dead ends" with the tell-tale symptom |
| A new technique worked better | Append it as a CANDIDATE rung with evidence |
| Same lesson now has 2+ occurrences | Promote: edit THIS SKILL.md (reorder/replace the rung), bump minor version, note the edit in LEARNINGS.md |
| Clean run, nothing new | Append nothing — no ritual entries |
Entry format (append to LEARNINGS.md):
## YYYY-MM-DD — <one-line summary>
- Symptom: <exact error / behavior>
- Rung affected: <0-3 / new>
- Evidence: <URL type, what was tried>
- Status: CANDIDATE | CONFIRMED (2+ runs) | APPLIED (SKILL.md edited vN)Known dead ends (do NOT retry these first — from real runs)
- claude-in-chrome extension on instagram.com: script injection times
out (
waited 45000ms for document_idle) — Instagram never goes idle. Affects screenshot, read_page, get_page_text. Don't burn retries. curlon/embed/captioned/: returns a 600KB JS shell with ZERO media URLs and no caption — the embed hydrates client-side. curl is fine for CDN image downloads (Rung 3), useless for the embed page itself.?__a=1&__d=disJSON endpoint: returns "Page Not Found" anonymously.- og-meta via WebFetch: works, but the "caption" is a paraphrased summary, not verbatim — never present it as a quote.
MUST DO
- Always read
LEARNINGS.mdbefore running the ladder — Why: the ladder self-corrects there first; skipping it repeats known-fixed failures - Always capture the caption VERBATIM from the embed snapshot before paraphrasing anything — Why: og-meta summaries have been mistaken for real captions; a quote must be character-exact
- Always verify downloaded slide file sizes before reading — Why: expired signed URLs produce tiny error bodies that waste visual-read calls
- Always include the Honest read section in the report — Why: the owner's standing rule requires assessment, not just extraction
- Always run STEP 5 (learning capture) — Why: this is the skill's self-improvement contract; a skipped capture is lost evidence
MUST NOT DO
- MUST NOT log in to Instagram or ask for credentials — the ladder is anonymous by design; if a rung demands login, record it and move on
- MUST NOT retry a "Known dead ends" path more than once — go to the next rung instead
- MUST NOT trust alt text for slide content — read the downloaded image
- MUST NOT edit SKILL.md on a single occurrence — record in LEARNINGS.md first; promote at 2+ occurrences (prevents one-off noise from rewriting a proven ladder)
- MUST NOT present engagement numbers without their fetch date
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

