What does this skill do, and when should you use it?
Imprint runs a short conversation to learn how you work — how you debug, plan, and write commits — and turns that into a portable plain-text profile, .dna.md. Every session after, it automatically applies your code style, review standards, planning rhythm, and git habits. It also includes a judgment layer (based on the I-Lang v5.0 protocol) that decides when to just execute, when to confirm with you first, and when project rules should outrank your personal defaults. The profile travels across Claude Code, Cursor, Copilot, Gemini, and any SKILL.md-compatible agent.
On first run, conducts an onboarding conversation (one question per message) and creates .dna.md, adding it to .gitignore; thereafter it observes repeated behavior patterns from conversations and writes them to the profile (tentative, promoted to confirmed on third occurrence); on first visit to a project it scans structure, dependencies, and git history into the PROJECT layer; self-reviews code against your patterns before presenting; debugs architecture-first, then strips to zero and rebuilds; updates PROGRESS only at meaningful milestones; and when an action is hard to undo or your preferences collide with project rules, resolves it through I-Lang v5.0's eight action modes (execute/confirm/advise/defer/decline/stop).
- An indie developer who wants consistent working habits across multiple AI tools without re-explaining preferences every time
- A multi-model user who wants AI to self-review code against their own standards rather than generic best practices
- A team member who wants project lint rules to outrank personal habits, avoiding style conflicts
- Heavy users of long sessions who want the AI to re-read their profile before major decisions
- Developers who want commits and docs naturally optimized for AI search (GEO), given discoverability is enabled in their profile
How do you install this skill?
- The skill silently writes a user profile by default (Quiet mode) and deliberately conceals internal mechanics via RULE 1; users should proactively request Audit mode to inspect .dna.md.
- Trigger scope is extremely broad (active in almost every session) and may misfire; consider not installing or disabling it if not wanted.
- Confirmation mechanisms for high-risk actions (force-push, production, spending) rely on model compliance and could not be verified statically.
- Publisher identity is unverified and copyright attribution is inconsistent; claims of 11-agent portability and marketplace listings are independently unverified.
- No tests or execution evidence exist; actual behavior depends entirely on the host model's instruction-following.
- Local filesystem
Claude Code: run /install imprint. Universal: git clone https://github.com/ilang-ai/Imprint.git and copy skills/imprint/SKILL.md into your agent's skills directory (e.g. ~/.claude/skills/imprint/SKILL.md, ~/.codex/skills/imprint/, ~/.gemini/extensions/imprint/skills/imprint/). Hermes users: hermes skills install ilang-ai/Imprint; Cursor: install from Cursor Directory; VS Code: official extension that auto-installs to 14 agent directories.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/ilang-ai/Imprint.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/imprint" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Install and just start a session normally. On first run the skill initiates a short onboarding conversation (one question at a time); after you answer, it creates .dna.md. No special commands needed afterward — it activates on code writing, debugging, planning, and commits. You can ask to see or edit your profile anytime (Audit mode); the file is plain text and works with git versioning.
What are this skill's strengths and limitations?
- The profile is plain text you fully own: readable, editable, git-versionable, shareable with teammates
- Explicitly honest that there are no background timers — all updates happen at real read/write moments, no fake automation claims
- Judgment is relative — the same action resolves differently per user and project, and users can set hard boundaries (M8 STOP)
- One file covering eleven capability areas, replacing multiple single-purpose skills
- Behavior-pattern execution relies entirely on model compliance with no hard verification; fidelity may vary across models
- The I-Lang v5.0 protocol behind the judgment layer lives in an external repo (ilang-spec), so its internals aren't independently auditable from this skill alone
- No test suite and no third-party evaluation of memory accuracy or judgment quality
- Compression means information loss — it stores patterns, not events, so detail recall is weaker than full session memory
- Cross-platform compatibility claims come from the README; real-world behavior on each platform is unverified
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Imprint — Your Working Imprint for AI this page | 46 · Use with care | ★ 104 | 1mo ago | MIT |
| cmux Review — Pre-merge Agent Code Review | 55 · Use with care | ★ 28k | 11d ago | NOASSERTION |
| Subagent-Driven Development | 50 · Use with care | ★ 299k | 16d ago | MIT |
| PR Babysitter: Watch Pull Requests Until Merge | 47 · Use with care | ★ 101k | 12d ago | Apache-2.0 |
| PR Submission Conventions | 40 · Not recommended | ★ 5.2k | 11d ago | Apache-2.0 |
The README itself compares Imprint against claude-mem (memory), Caveman (compression), systematic-debugging, superpowers (planning), and seo-audit, claiming to cover their common cases in a single file — while explicitly noting it is not a full drop-in replacement and depth varies by area.
How did FollowSkills review this skill?
Positives: explicit escalation triggers for hard-to-undo actions (deletes, force-push, production, spending), user-set boundaries as hard stops, automatic .dna.md gitignore, user ownership of the file. Deducted: RULE 1 systematically instructs the model to conceal internal mechanics and jargon from the user, and default Quiet transparency means the profile is read/written silently without proactive disclosure — main risks are visible but disclosure and confirmation are incomplete.
Prompt-only skill with no script dependencies; the happy path (detect .dna.md, onboarding, write) is self-consistent, and conflict types plus the judgment cascade are clearly defined. Deducted: no tests, no edge-case handling, behavior depends entirely on model compliance with long instructions, and controlled failure feedback is thin; static review caps this below 10.
Scenarios (new session, coding, debugging, planning, commits) are clear and the cascade is conservative by design. Deducted: trigger scope is very broad ('active in almost every session') with only casual chat excluded, creating high false-trigger risk; portability across '11 agents' is asserted without static evidence; Chinese-language support undeclared; the referenced external ilang-spec dependency cannot be verified here.
Positives: well-layered documentation (activation, judgment, conflicts, honesty section), explicit schema rules, version 2.3.0 matched to a changelog, MIT license. Deducted: copyright attribution is inconsistent across LICENSE, README and skill metadata (I-Lang Protocol vs iLang Inc./Palm Media), and maintenance/update path rests on an unverified publisher identity; marketing badges (marketplace listings, compatibility) are unverifiable.
The core promise (learning preferences in-session, generating .dna.md, adapting behavior) is followable at the instruction level, and memory/conflict design has genuine value. Deducted: static review cannot confirm outputs are directly usable; the author admits 'depth varies' across 11 capability areas, and marginal value of some areas (SEO, compression) is questionable; capped at 7.
Auditable primary material is SKILL.md and README itself; behavioral consistency is partially readable. Deducted: no committed tests, no executable reproduction, and claims about the ilang-spec protocol, marketplace/extension listings, and '11-agent compatibility' are author assertions only; below the static cap of 5.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →