instagram-list-inspector
Analyze user-accessible Instagram followers, following, and post/Reels likers and commenters in a signed-in Codex or Claude browser; compare audiences and content topics, track dated changes, and create evidence-aware interactive reports.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/greedyc--instagram-list-inspector)HTML
<a href="/listings/greedyc--instagram-list-inspector"><img src="/api/listings/greedyc--instagram-list-inspector/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Clone this repository into your personal Codex skills directory so that `SKILL.md` is inside the `instagram-list-inspector` folder:
git clone https://github.com/GreedyC/instagram-list-inspector.git "$HOME/.codex/skills/instagram-list-inspector"
For Claude Code, clone into `~/.claude/skills/instagram-list-inspector` instead. The shared entrypoint routes separately for each client. Restart or refresh skill discovery if needed, then ask for a specific comparison, for example:
> `$instagram-list-inspector` Compare the accounts followed by @account_a and @account_b. Show the shared handles and tell me whether both lists were complete.
The browser may request login, verification, or CDP approval. Complete those prompts yourself. The skill does not ask you to reveal credentials. If browser/CDP access is unavailable, it reports the limitation instead of claiming a complete result.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
SKILL.md · Manifest · 64 lines
- ---
- name: instagram-list-inspector
- description: Analyze user-accessible Instagram followers, following, and post/Reels likers and commenters in a signed-in Codex or Claude browser; compare audiences and content topics, track dated changes, and create evidence-aware interactive reports.
- ---
- # Instagram List Inspector (Codex and Claude)
- Handle the analysis in this task, without the Followloom CLI, a ZIP export, pasted lists, or a copied session cookie. The user supplies only target handles/post URLs and the question. For operation definitions, the snapshot contract, and report rules, read [analysis-playbook.md](references/analysis-playbook.md) before analyzing or saving data.
- ## First-use orientation
- On an explicit invocation such as `$instagram-list-inspector`, give a short orientation **before browser actions**, in the user's language. Cover what the user can ask, the minimum input, the only possible handoffs, and the default output. Keep it conversational, not a long questionnaire. If the invocation contains no concrete target/task, briefly cover these capabilities and ask one question:
- - Mutual following and non-follow-backs for one account.
- - Shared or distinct followers/following across accounts.
- - Overlap between post/Reels likers and account audiences; repeat likers across posts.
- - Compare user-labelled content topics and observed commenters × post likers × audiences.
- - Dated additions/removals when earlier captures exist.
- - In-chat findings by default, or a requested local HTML/CSV/JSON report.
- Ask for only the target handle(s) or post/Reels URL(s) and the desired comparison. Do not ask for a password, cookie, ZIP export, or pasted lists. The user may need to complete Instagram login/verification or approve the browser's CDP access. A historical comparison needs an earlier capture; if none exists, offer to create a baseline. In Turkish, a concise friendly opening is appropriate; in other languages, match the user's tone.
- If the user already supplied a complete request, condense this to one sentence (what will be compared and that no credential/export is needed), then proceed without asking them to pick from a menu. If only the target or the intended comparison is missing, ask for that one missing piece. Do not start collecting an unrelated account merely to demonstrate the skill. Do not promise a complete list before access and coverage are checked.
- ## Scope
- Support all analytical features from the Followloom product plan when the necessary lists are accessible:
- - One account: followers ∩ following (mutuals), followed-but-not-following-back, followers-not-followed-back.
- - Two or more accounts: shared/unique followers or following, pairwise and three-way intersections, overlap with explicitly named denominators, and candidate audience bridges.
- - A post or Reel: observed likers ∩ any selected account's followers/following; observed likers outside those lists; multiple posts' repeat-liker and campaign-overlap analysis.
- - Dated captures: additions/removals between comparable snapshots, changes in intersections, and careful treatment of username changes or disappeared accounts.
- - Creator analyses: user-labelled post topics, distinct/repeat observed participants across posts, commenters × post likers × selected audiences, and common questions/themes when the user requests analysis of observed comment text. Read [creator-analysis.md](references/creator-analysis.md) for comment collection, exact counting and the deterministic helper.
- - Evidence-aware output: coverage and source for every set, partial-data sensitivity, concise in-chat findings, and user-requested local JSON/CSV/HTML reports with aggregate-only or pseudonymous sharing options.
- Do not confuse supported *analysis* with guaranteed data access. Instagram may withhold a list, limit a response, change a private web endpoint, or challenge the session. Never fill unavailable data by guessing, and never claim a negative observation from a partial list.
- ## Client routing
- Read the actual browser-tool documentation before collection. Skill installation does not grant a capability.
- - Codex: use the documented browser interface. Follow the CDP path below only if the tool explicitly exposes and permits session network inspection and same-origin requests. If it does not, use visible UI observation and report coverage limits.
- - Claude: use connected Claude in Chrome, Cowork browser, or computer-use tools. Check the current session's tools first; `claude --chrome` and `/chrome` are setup/status options for Claude Code. Default to visible UI and accessible page text. Do not assume Codex APIs, CDP, or network interception exist in Claude.
- - Visible UI path: read rows before each scroll, deduplicate observed handles, record displayed totals and stopping reasons. No change after scrolling does not prove completeness. Stop on challenges, access denials, rate limits or recycled rows. Use the same analysis and reporting contract below.
- - Only use a network path in either client when the current tool documentation explicitly permits it. Do not add external browser servers or enable integrations automatically.
- Claude capability references: [Chrome](https://code.claude.com/docs/en/chrome), [computer use](https://code.claude.com/docs/en/computer-use). Availability is checked at runtime.
- ## Collect with minimum user effort
- 1. Resolve the requested targets from the user's message. Reuse a signed-in Instagram tab through the browser tool; ask only for missing handles/post URLs that cannot be inferred. The user handles login, verification, and any full-CDP approval. If browser/CDP access is unavailable, explain it; use visible UI observation when supported and describe its coverage limits.
- 2. For visible computer use, read each visible row before scrolling, use the normal list/reply controls, and retain source/scope/stop evidence; skip the network-only instructions in steps 2–3. For commenters, distinguish comment authors from mentions and post likes from comment likes; read [creator-analysis.md](references/creator-analysis.md). Only when the tool explicitly permits a network route, follow its initialization and CDP capability documentation. Observe network events around opening the requested list once in the normal UI. Identify the actual same-origin list request and response; never assume an endpoint from an older run. Read only the fields needed for ID/handle matching, displayed/returned totals, and pagination.
- 3. If more pages are advertised, use the *same tab's existing session* for sequential, bounded, same-origin requests derived from that observed GET and its cursor. Keep necessary headers and tokens in memory only. Do not print, persist, export, or pass cookies, CSRF tokens, raw request URLs, or full network bodies to another service. Stop at a challenge, 429, access denial, cursor loop, malformed/empty page, or reasonable page cap; no bypass, proxy rotation, or aggressive retry.
- 4. For each list record its profile/post URL, kind, UTC capture time, displayed or response total and its source, observed identity keys/current handles, pages read, pagination-end evidence, access/stop reason, and coverage status. Prefer stable IDs; without them use case-insensitive handles and mark identity provisional. If pagination ends short of a credible total, including handle-only captures, follow the bounded [count-reconciliation procedure](references/analysis-playbook.md#count-reconciliation). Preserve both passes separately in the version-2 snapshot contract. Neither a matching count alone nor provisional identity establishes full coverage.
- Reinspect each list type's live request and response before relying on it. A successful capture from another session is not evidence that a new capture is complete.
- ## Analyze and deliver
- Use [analysis-playbook.md](references/analysis-playbook.md) for exact set definitions, temporal conditions, partial-data language, report schema, and privacy defaults. Compute only from observed IDs/handles. Label every table/claim with capture times and coverage. Say “not observed in this capture” rather than “did not like,” “unfollowed,” or “has a relationship.” For a historical comparison, save dated local snapshots only when the user requests tracking or a report that requires them; tell the user the local path and do not put account lists in vault memory or a public repository.
- Return the answer in chat by default. Generate local HTML/CSV/JSON only when requested; default any shareable report to aggregate results, with identities included only when the user asks. No auto-publishing, scheduled monitoring, mobile app, or CLI is implied by this skill.
- For the selected creator features, keep collection in this agent/browser workflow and use the standard-library helper in [creator-analysis.md](references/creator-analysis.md) for repeatable calculations. It does not collect data or require a separate agent/service. A clickable HTML report is opt-in; its export defaults to aggregates, and filters clearly separate the post table from whole-report topic totals.
README.md · README · 67 lines
- # Instagram List Inspector
- An open-source **Codex and Claude skill** for comparing Instagram lists accessible in your signed-in browser. It runs on demand and reports observed data and coverage limits.
- The skill can compare followers and following across accounts, find mutuals and non-follow-backs, intersect post/Reels likers with account audiences, spot repeat likers across posts, compare dated captures, and analyze commenters alongside post likers and user-labelled content topics. It joins accounts by platform ID where available and tells you when a list is incomplete. Results appear in chat by default; local JSON, CSV, or HTML reports are optional.
- ## Requirements
- - Codex or Claude with an available signed-in browser/computer tool. Codex uses CDP only when the tool explicitly supports it; Claude defaults to visible UI observation.
- - For Claude Code, see [Chrome integration](https://code.claude.com/docs/en/chrome): launch with `claude --chrome` and check `/chrome`. Installing this skill does not activate browser tools.
- - An Instagram account with legitimate access to the specific lists you ask to inspect.
- - Your own judgment about applicable Instagram terms, privacy rules, and permissions. No password, session-cookie copy, ZIP export, or CLI setup is required by the skill.
- ## Install and use
- Clone this repository into your personal Codex skills directory so that `SKILL.md` is inside the `instagram-list-inspector` folder:
- ```bash
- git clone https://github.com/GreedyC/instagram-list-inspector.git "$HOME/.codex/skills/instagram-list-inspector"
- ```
- For Claude Code, clone into `~/.claude/skills/instagram-list-inspector` instead. The shared entrypoint routes separately for each client. Restart or refresh skill discovery if needed, then ask for a specific comparison, for example:
- > `$instagram-list-inspector` Compare the accounts followed by @account_a and @account_b. Show the shared handles and tell me whether both lists were complete.
- The browser may request login, verification, or CDP approval. Complete those prompts yourself. The skill does not ask you to reveal credentials. If browser/CDP access is unavailable, it reports the limitation instead of claiming a complete result.
- ## Privacy and limitations
- - The workflow observes visible Instagram UI by default. Unofficial web requests are an optional route only when the current browser tool explicitly permits it; either route can change or stop working. It does not claim Meta/Instagram affiliation or approval.
- - Meta says automated data collection without permission can violate its terms. Use only where you have appropriate rights and permission; respect platform limits and stop on challenges or rate limits. See [Meta's scraping guidance](https://www.facebook.com/help/463983701520800) and the [Instagram Terms of Use](https://help.instagram.com/581066165581870).
- - Private or unavailable lists are not bypassed. Missing results from partial captures are **not** proof that a person did not follow or like something.
- - Instagram may end pagination before the observed identity-key count (IDs, or normalized handles when IDs are unavailable) matches a profile total. The skill makes at most one bounded consistency pass and preserves each pass's evidence separately. Handle-only identity stays provisional even when counts match; unresolved mismatches remain partial.
- - Session headers stay in browser memory during a run. Never publish cookies, raw responses, screenshots containing private information, or generated reports. The repository contains skill instructions, a local analysis helper and synthetic fixtures, not real account data.
- - Historical comparisons require an earlier user-requested local snapshot. The skill does not start background monitoring.
- This project is licensed under MIT; that license does not grant rights to Instagram data or waive platform terms.
- ## Türkçe kısa rehber
- Bu proje, Codex içinde çalışan bir **skill**: erişebildiğin Instagram takipçi/takip edilen ve gönderi-Reels beğenen listelerini karşılaştırır. Ortak hesapları, karşılıklı takibi, geri takip etmeyenleri ve tarihli kayıtlar arasındaki değişimleri gösterebilir. Eksik veri varsa bunu açıkça söyler; “listede görünmedi” sonucunu “kesinlikle yok” diye sunmaz.
- Kurulum için yukarıdaki `git clone` komutunu kullan. Ardından örneğin şunu yaz:
- > `$instagram-list-inspector` @hesap_a ile @hesap_b'nin ortak takip ettiklerini bul; iki listenin de kapsamını belirt.
- Şifre, çerez, ZIP veya elle liste yapıştırman gerekmez. Codex veya Claude içinde kullanılabilir tarayıcı/computer-use aracı gerekir; CDP zorunlu değildir. Bu, resmi Instagram API'si değildir; platform kurallarını ve kişilerin gizliliğini gözeterek kullan.
- ## Creator analysis and clickable reports
- Ask the agent to compare posts by topic (for example, Coffee vs Coding), cross
- observed commenters with post likers/followers, or summarize questions in comments
- it can actually observe. The agent uses its available browser/computer tools; a
- separate agent, collector service or user-run CLI is not required.
- A Python 3.10+ standard-library helper supplies repeatable ID-based calculations
- and requested self-contained HTML reports. The report has post search, topic and
- capture-date filters, sorting and inline source evidence; it runs offline. Whole-
- report topic counts remain separate from post-table filters. Participant identities
- in membership lists are omitted from aggregate exports; requested source URLs and descriptive metadata remain. The agent must keep descriptive fields neutral for sharing. Reports support optional report-local pseudonyms or
- explicitly requested identified reports. No third-party Python packages are needed.
- See [creator-analysis.md](references/creator-analysis.md) for the normalized input,
- comment-scope rules and agent workflow. New features are verified with synthetic
- fixtures; that is not a claim that Instagram currently exposes every requested
- list or that live multi-account collection has been validated.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.