core

Plugin core from marketplace ibaad-toolkit in ibaadrh/agent-toolkit. Commands read from the repo's manifest — 19 days ago.

Skills (19)

• ai-image-prompting

  Rules for writing prompts for AI image and video generators (Firefly, Nano Banana Pro, Veo and similar) so the result looks like real photography and any real product in it stays exact. Covers composition zones, the fixed product block, naming generic devices, people, light and lens, the ban list, r

• brief-contract

  The exact `## BRIEF` block that the ads strategist writes and the ad makers and campaign operator parse, plus how each maker uses every field and what each one hands back. Use it when writing a brief, when a skill receives a `## BRIEF` block, when checking a brief before hand-off, or when building a

• browser-automation-hygiene

  House rules for any skill that drives a real browser: wake a page in a hidden pane with one screenshot, retake a screenshot that may still show the previous state, and never trigger a native dialog (move files in by in-page fetch of a public URL instead). Use it whenever a skill navigates, reads or

• claims-and-voice-rules

  The truthfulness rules every client gets (never invent facts, approved claims only, no fake testimonials, AI presenters never framed as customers, awards in their exact wording, no capability the product lacks), the author's default copy rules (no em dashes, write the number not the adjective, "as m

• client-config

  How every toolkit skill finds and reads the client's config files (brand.yaml, ads.yaml, meta.yaml, voice.yaml, product-spec.md and the rest), what counts as missing, and what to do then. Read it at the start of any skill run that needs client details, whenever a config file or key cannot be found,

• ffmpeg-vertical-toolkit

  Tested ffmpeg recipes and checks for vertical (9:16) video, for ads, reels and UGC-style edits. Covers 9:16 crop and blur-background framing, frame-accurate trims joined by concat with re-encode, trimming the end of AI clips, the loudnorm audio chain, ASS captions inside Reels and Stories safe zones

• file-delivery

  How to hand finished files to the user and put them onto their machine or cloud folder safely. Covers naming and versions, the per-file size limit and splitting large files, never deleting (move to a scratch folder instead), checking that every file landed by listing the folder, recovering from a pa

• firefly-web-driver

  Driving Adobe Firefly (firefly.adobe.com) from a browser the agent controls, for image generation (Nano Banana Pro and other models), video generation from a first frame (Veo), and Firefly Speech voice-overs. Covers the in-page helper set, the desktop viewport, the 45-second script limit, reference

• house-report

  The one report format every toolkit skill ends with. One sentence on what was decided or built, two or three metrics, what changed, then numbered points for the owner or the line "Nothing needed from your side." Use it at the end of every run, when writing a run-log line, and whenever a skill says "

• llm-council

  Run any question, idea, or decision through a council of 5 AI advisors who independently analyze it, peer-review each other anonymously, and synthesize a final verdict. Based on Karpathy's LLM Council methodology. MANDATORY TRIGGERS: 'council this', 'run the council', 'war room this', 'pressure-test

• meta-creative-publish

  The safe way to turn a finished image or video into a live Meta ad. Covers the RGB and size check, uploading by public URL to get an image hash, why creatives are immutable, creating the ad paused, previewing, activating with explicit IDs, verifying the effective status, and AI self-disclosure. Use

• park-a-note

  Append one dated line to the project's session notes file so an idea, a to-do or a loose end is not lost, without stopping the current work or ending the session. Use it when the user says "park this", "note that for later", "remind me to", "add to the list", or mentions something worth keeping that

• platform-adapters

  Maps the actions that toolkit skills name (run JavaScript in the browser page, deliver the file to the user, search the Meta Ads Library, ask the user a question) to the tool names of the agent you are running in. Read it whenever a skill step names an action and you do not know which of your tools

• product-accuracy-check

  The gate every AI image or 3D render with a real physical product must pass before design work starts. A measured silhouette check with foreshortening correction, a feature checklist driven by product-spec.md, cable and connector physics, which end is which, the brand look-alike gate, honest retouch

• readable-test-doctrine

  The rules that decide whether a paid-ads test can be read at all, and how to build one that can. Covers the readability gates (impressions per cell, leader link clicks), one ABO ad set per cell and never a split test inside CBO, the spend-share check, the decision-metric order, the rule-of-three tab

• resume-session

  Start a session on an existing project by reading its state and the last session notes, then saying in a few lines where things stand and what the next step is, before doing any work. Use it when the user says "resume", "where were we", "pick up where we left off", "what's next on this", or opens a

• shell-safety-guard

  A check that stops shell commands which break, or silently change meaning, when the user's shell is zsh, and says the fix for each. It catches words starting with "=" ("echo ====", "[ a == b ]"), unquoted globs inside flags ("grep --include=*.md"), after "=" in arguments, as the argument of pattern

• shopify-cdn-publish

  Host a file at a public URL when the agent cannot put it on the web directly, using a Shopify store's Files as the host. The four-step staged upload (ask for an upload slot, POST the file, register it, wait until it is ready) and why it is needed. Use it when an ad platform, an image generator in a

• wrap-up-session

  End a session by appending a short note to the project's session notes file, saying what was done, what is still open and the one next step, written so a cold session (or another person or agent) can pick it up. Use it when the user says "wrap up", "let's stop here", "write it down for next time", "

Slash commands (3)

• /core:park-a-note

  Append one dated line to the session notes without stopping the current work

• /core:resume-session

  Read the project's state and last session notes, say where we are and what's next

• /core:wrap-up-session

  Append a short session note (done, open, next step) to the project's notes file

Hooks (1)

• PreToolUse

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/zsh-guard.sh"