agentleFS
uplikauplika/uplika-plugin/skills/ig-caption/SKILL.md

ig-caption

Skill3 starsChanged 34 days agoMITView on GitHub

Write and lint an Instagram caption (the line before the more cut, body, one ask, search terms, a few hashtags). Use for captions and hashtags.

11 other projects use a version of this, 7 of them edited.

Adopted by

Most recently taken up by

Used with

Projects with it also use ig-plan and ig-reply.

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.
  • JakeschincariolSkill

    ig-plan

    Build the week on Instagram - what to post, which format, when, and who to engage with. Use when the user says "plan my week", "what should I post", "content calendar", "I have nothing to post about", or wants a posting schedule and an engagement list.

    Also in 8 projects that use this

  • vava2684Skill

    ig-reply

    Instagram 貼文留言智慧回覆。抓某篇貼文底下的新留言,由 Claude 在對話裡依貼文內容與留言者語言生成友善回覆,再透過 Graph API 自動送出。觸發時機:使用者說「回覆留言」「處理 IG 留言」「回 IG」「IG 自動回覆」「抓留言來回」「回我 IG 那篇」等。

    Also in 7 projects that use this

The file

We split this skill into the rules it gives an agent

9 rules. 3 are common, 1 is distinctive, and 5 are very specific to this project.

  • 3

    Common

    Most of the 11 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use when the person says "write the caption", "caption this", "what do I put in the description", has a reel or a carousel ready and needs …

    line 10
  2. Very specific:

    Read the voice file ~/.uplika/<workspace-slug>/voice.md first if it exists.

    line 23
  3. Common:

    Ask which one you are writing.

    line 33
  4. Common:

    Ask for two or three of those terms, then pass them to the linter:

    line 62
  5. Common:

    Write the first comment separately: if there is a link. Say so in the receipt.

    line 73
  6. Very specific:

    Run ig-human on it. Captions are short, so slop is louder here than anywhere else.

    line 81
  7. Very specific:

    Run scripts/caption.py with the person's search terms.

    line 82
  8. Very specific:

    If a call answers workspace_required, the account has more than one workspace: show the names it lists, ask which one, and pass that …

    line 97
  9. Very specific:

    Check the length against Instagram in list_platforms (its character limit and charCount rule).

    line 101

“Common” is measured: kept by at least 6 in 10 of the 11 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Adapted from Jakeschincariol/instagram-agent-skill (MIT). Not affiliated with the original author.

ig-caption

Use when the person says "write the caption", "caption this", "what do I put in the description", has a reel or a carousel ready and needs the text, or asks about hashtags.

One script lives in this skill's scripts/ folder and it runs. From this skill's folder:

python3 scripts/caption.py caption.txt
python3 scripts/caption.py caption.txt --keywords "client proposals,agency pricing"

It prints the caption the way the feed prints it: the part shown before "... more" in a box, everything else behind the tap. That cut-off is an approximation that moves with the device and the font size, and --truncate tests a tighter one. Read that box before you read anything else you wrote. The script exits non-zero until the verdict is READY; that is its answer, not a crash.

On Korean captions the length, first-line and search-term checks still work. The hashtag counter only sees tags written in Latin letters, and the ask and concreteness checks look for English patterns (a number written together with a Korean unit is not counted), so check those by reading. Without a terminal the script cannot run: apply the checks by reading and say the lint was not run.

Read the voice file ~/.uplika/<workspace-slug>/voice.md first if it exists. If it does not, ask three questions (tone, how they end a line, words they avoid) and offer to save the answers with the voice-profile skill.

First, decide which job this caption has

This is the decision that ruins captions when it is skipped.

Job A: the video already hooked them. A Reel carries its own hook in the first two seconds, spoken and on screen. The caption is not a second hook and competing with the video is how you lose both. Its job is the ask, the context that makes the ask make sense, and the words people search.

Job B: the caption is the content. A photo, a single image, a carousel cover that opens a loop. Here line one is the hook and it works exactly like a Reel hook: concrete, short, and cut off at a cliff rather than mid-clause.

Ask which one you are writing. If the person has a Reel with a strong hook, write A and say why.

The shape

Line 1      the space the feed shows before "... more". Job A: the ask,
            plainly. Job B: the hook.
            Never a greeting, never a hashtag, never an emoji as the first
            character.
Body        short paragraphs, one line of white space between each. Two to six
            of them. This is where the search terms live.
The ask     one. Comment a keyword, save it, or DM. One.
Hashtags    a few, on their own line at the bottom, or none.

How long a caption may be, and how Instagram counts its characters, comes from list_platforms. Almost nothing needs the whole limit. A caption that earns the tap and then delivers a few tight paragraphs beats one that fills the box.

Hashtags, honestly

Hashtags are not a reach lever any more, and the platform has now said so with a product change. In December 2025 Instagram cut how many hashtags a post can carry, telling creators that fewer, more targeted hashtags perform better than many generic ones. Adam Mosseri had already said in February 2025 that hashtags do not work to increase reach and are a label, not a distribution lever.

So: a few, specific, as topic labels. If the person has a long block saved in their notes app, that block is now dead weight and the linter will fail it. scripts/caption.py carries the cap the original author checked when writing it. Uplika's list_platforms does not carry a hashtag cap, so if Instagram's own guidance has moved since, Instagram is right.

#viral, #fyp, #explorepage, #foryou describe nothing. Cut them.

Search terms matter more than hashtags now

Instagram search reads the caption text. So the phrase the person wants to be found for goes in the caption as a phrase a human would type, in a sentence that reads normally. "Client proposals" as words in line three, not "#clientproposals" in a block at the bottom.

Ask for two or three of those terms, then pass them to the linter:

python3 scripts/caption.py draft.txt --keywords "client proposals,agency pricing"

Rules

  • No link in the caption. Captions are not clickable. A URL in the body is dead text that says "I do not use this platform". Bio or DM.
  • One ask. Two asks is the same as none. scripts/caption.py counts them.
  • The keyword ask needs a keyword people can type. One word, no spaces, no emoji, and say it out loud in the video too. Comment CONTRACT works. Comment "the contract guide" does not.
  • Write the first comment separately if there is a link. Say so in the receipt.
  • Emoji as punctuation, not decoration. The linter flags heavy emoji use.
  • Alt text is worth 20 seconds. For carousels and photos, write it. It is read by screen readers and by Instagram.

The loop

  1. Decide Job A or Job B and say which.
  2. Draft it.
  3. Run ig-human on it. Captions are short, so slop is louder here than anywhere else.
  4. Run scripts/caption.py with the person's search terms. Fix every FAIL. Decide on every WARN out loud rather than silently.
  5. Print the copy-ready block, then the receipt:
CAPTION READY
job:        A - the reel carries the hook
visible:    118 characters used before the cut
ask:        one, comment CONTRACT
hashtags:   3
search:     "client proposals" in line 3, "agency pricing" in line 5
linter:     READY

With Uplika connected

If a call answers workspace_required, the account has more than one workspace: show the names it lists, ask which one, and pass that workspaceId on every later call in this session.

This skill writes the caption and does not publish. The caption becomes the content of the post that ig-reel, ig-carousel or the publish skill sends, after the person approves it there.

  • Check the length against Instagram in list_platforms (its character limit and charCount rule). Emoji can count differently there than in scripts/caption.py; when the two disagree, list_platforms wins.
  • Instagram captions cannot be edited after the post goes out (the features table in list_platforms says so). The linted, approved text is the final text.
  • The separate first comment goes in options.instagram.firstComment. Uplika posts it right after the post (not on stories), and if it fails the post still goes up with a warning.
  • Stories show no caption; their words go on the frames (ig-story).
  • A keyword ask ("Comment CONTRACT") needs the automation that sends the thing: the automations skill sets up template:comment_to_dm. Use the exact same keyword in both places.
  • Alt text travels with each image when it is uploaded (altText), not in the caption.

Without Uplika

Print the block and the receipt. The person pastes the caption into Instagram.

Tools

  • list_platforms: Instagram's caption limit and counting rule, the option schema, and the features table (no edits after publishing)

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.