personas

claude-personas

Session-aware persona advisors for Claude Code. Each persona (Steve Jobs, Taylor Otwell, Raymond Hettinger, David Tolnay, or any you author yourself) becomes a long-lived voice with project-scoped threads and durable memory. Speak naturally: "ask steve what he thinks", "what would taylor say about this controller?" Threads persist across CC sessions; closed threads distill into a per-project memory file the persona reloads next time.

What this is for

Stop re-introducing yourself. The standalone *-says plugins were stateless: one prompt, one answer, no continuity. The personas plugin turns each persona into an advisor that remembers what you discussed last week, what advice you took, and what's specific to this project versus the one in another folder.

Project-scoped heads. Same Steve, different head per project. Steve in your Laravel app remembers your launch plan; Steve in your Rust crate remembers the API decisions. Project identity is keyed off your git remote, so worktrees and clones share state.

One plugin, many personas. Install claude-personas once. Bundled personas ship disabled by default. Opt in with /personas enable steve-jobs. Add third-party personas with /personas add <git-url>; they're plain repos with a persona.md, no Claude plugin scaffolding required.

Natural-language management. A personas skill detects "ask steve about X" without slash commands, continues open threads automatically, and silently auto-closes resolved threads. Closing distills the persona's takeaways into the memory file before archiving the transcript. /personas reopen <name> is the safety valve.

Installation

# Clone into your plugin directory (until published to a marketplace)
git clone https://github.com/mischasigtermans/claude-personas

# Or install via your plugin manager
/plugin install mischasigtermans/claude-personas

Quick start

# See what's available
/personas list

# Enable a bundled persona
/personas enable steve-jobs

# Then just speak
"what would steve think about removing this feature?"
"ask taylor whether this controller is doing too much"
"raymond, is there a more pythonic way to write this?"

You can also be explicit:

/personas ask steve-jobs "Evaluate this product brief..."
/personas threads               # show open threads in this project
/personas close steve-jobs      # write takeaways, archive thread
/personas reopen steve-jobs     # restore the most recent closed thread

Bundled personas

Canonical name	Aliases	Domain
steve-jobs	steve, jobs	Product design, leadership, simplicity, craft
taylor-otwell	taylor, otwell	Laravel reviews, code elegance
raymond-hettinger	raymond, hettinger	Pythonic style, code reviews
david-tolnay	david, tolnay	Rust idioms, type safety

All ship disabled by default. Each bundles its own knowledge files (deep guidance per topic) loaded on demand.

How it works

┌─ skills/personas/           ◀── triggers, dispatch, lifecycle
│                                  awareness ("ask steve...") + /personas actions
├─ personas/<name>/           ◀── bundled persona definitions
│   ├─ persona.md             (YAML frontmatter + system prompt)
│   ├─ context/               (voice, quotes, personality. Inlined via @)
│   └─ knowledge/             (deep topical guidance. Read on demand)
├─ ~/.claude/personas/
│   ├─ config.json            (enabled set)
│   ├─ external/<name>/       (third-party personas you /personas add)
│   └─ state/<project_id>/<persona>/
│       ├─ memory.md          (durable, distilled)
│       ├─ open-thread.json   (pointer if a thread is active)
│       └─ threads/<id>.md    (full transcripts)

project_id is sha1 of the git remote URL (first 12 chars), with cwd hash as fallback. State lives outside the plugin and survives reinstalls.

Threads vs memory

• Thread (active). Full transcript, replayed in the persona's context on every continuing turn.
• Memory (durable). Short bullet list of facts, preferences, and decisions the persona "knows" about this project. Loaded as preamble on every new thread.

When a thread closes (manually or via auto-close heuristic), the persona writes 3–8 bullet takeaways to memory.md, then archives the transcript. Reopening a closed thread restores it as the active thread without removing the takeaways.

Silent auto-close

The skill closes threads on its own when:

• Idle: 3 consecutive user turns without involving the persona.
• Topic shift / closure cue: user signals "thanks", "got it", "anyway", or moves to an unrelated topic.
• Resolution: the persona's last reply gave a clear answer and the user moved on.

You'll see a one-line notification: "Closed steve-jobs thread (idle). /personas reopen steve-jobs to restore." The transcript and takeaways are preserved either way.

Authoring a persona