| name | emoji-consistency |
| description | Keep emoji usage consistent across sessions. Whenever Claude is about to use an emoji to represent a project, task, subtask, or category in a user-facing message (Telegram reply, commit message, status update, TODO list, PR title, progress report, etc.), look the name up in ~/.claude/emoji-assignments.json first and reuse the same emoji. If unknown, pick a fitting emoji and record it. Triggers on phrases like "assign emoji", "emoji for project", "what emoji", "consistent emoji", and implicitly on every emoji-in-message decision for a named entity. |
| user-invocable | true |
| allowed-tools | ["Read","Bash(*)"] |
Emoji Consistency
Use this skill to keep entity emoji stable across Claude Code sessions. The same project, task, subtask, or category should get the same emoji every time, even when the decision was made in an older conversation.
Invoke it whenever Claude is about to write an emoji that represents a named entity in a user-facing message, including Telegram replies, commit messages, status updates, TODO lists, PR titles, and progress reports. Do not invoke it for decorative or generic status emoji that do not represent a specific named entity, such as checkmarks, celebration marks, or warning symbols.
Workflow
-
Before emitting an emoji for entity X, run:
~/.claude/skills/emoji-consistency/scripts/lookup.sh "X"
If it exits 0, use the printed emoji.
-
If lookup exits 1, pick a creative, fitting emoji yourself, then record it:
~/.claude/skills/emoji-consistency/scripts/record.sh "X" "🎚" project
Use the printed emoji from record.sh in the final message.
-
Prefer specific, memorable choices. A music production app is 🎚, not 🎵; a voice assistant is 🦜, not 🔊; a Telegram bot is 🛰, not 💬. Avoid reusing the same emoji for different entities unless the user explicitly wants that.
Kind Classification
project: top-level work, such as producer-player, agent-bridge, or agentlicationtask: a named workstream inside a project, such as producer-player:mastering-pipeline or agent-bridge:tailscale-migrationsubtask: a smaller unit inside a taskcategory: a cross-cutting topic, such as bug, feature, docs, or release
For tasks and subtasks, pass the parent project when it is known:
~/.claude/skills/emoji-consistency/scripts/record.sh "mastering pipeline" "🎛" task "producer-player"
Lookup Notes
For exact lookups, always use lookup.sh; it normalizes names and updates dateLastUsed.
For bulk or loose scans, such as checking which projects already have assigned emoji, rg '"name":' ~/.claude/emoji-assignments.json is fast enough and parallel-safe for reads. Do not use ripgrep for exact lookup or write decisions; use the scripts.
Pruning
The scripts automatically prune old entries in the background when the JSON file grows past 200 entries or 50 KB. Subtasks expire after 30 days, tasks after 90 days, and categories after 180 days. Projects are never auto-removed.
The prune.sh --force flag only skips the auto-prune scheduling gate; age rules still apply and projects remain sacrosanct. If project deletion is added later, children with matching parent should be cascade-removed.
Storage Design
Assignments live in a single JSON file at ~/.claude/emoji-assignments.json. This is intentionally simple: 1000 entries at roughly 150 bytes each is about 150 KB, which remains trivial to parse with jq and inspect with ripgrep. No sharding is needed until around 10k entries.
Example
emoji=$(~/.claude/skills/emoji-consistency/scripts/lookup.sh "producer-player") || {
emoji=$(~/.claude/skills/emoji-consistency/scripts/record.sh "producer-player" "🎚" project)
}
echo "$emoji producer-player"
