cairn-build-user-profile

cairn: Build-user-profile skill

Maintain the user profile — a living synthesis of how the user thinks, decides, and collaborates. Not what they prefer at a single-issue level (those are individual memory entries); how they reason at the meta level.

When to use this skill

• The user reveals a durable preference about how they work (editor, license, autonomy comfort, decision style).
• The user corrects your framing in a way that reveals a pattern ("too narrow," "too formal," "pick one visibly").
• The user confirms an unusual choice without pushback — that's a validated judgement worth capturing.
• At session close (invoked from close-session skill) — check for new signal worth synthesising in.

Do NOT use for

• Task-ephemeral preferences (which file to edit now).
• Frustration, impatience, or emotional reactions — this is about working patterns, not moods.
• Project-specific stances (use build-project-profile).
• Random one-off observations — need at least two data points or an explicit statement before it goes in.

Profile location (resolve at start)

In order:

1. $CAIRN_USER_PROFILE env var → use that path.
2. <project>/.cairn/config.json → if present and has user_profile_path, use that.
3. ~/.config/cairn/user-profile.md → default (global).

If none of the above exists and the user-profile feature is enabled, offer to bootstrap at the global default and prompt for the install-time location choice. If the user skipped the feature entirely (option 4 at install), don't silently create one — ask.

How to update

Observations with evidence, not judgments.

Every new entry needs:

• A specific section (Communication style / Decision-making pattern / Design values / Collaboration posture / Technical preferences / Open tensions).
• A one-line observation.
• Evidence — a session date, conversation reference, or a quote from the user. Evidence makes the observation refutable, which keeps the profile honest.

Example:

## Communication style

- **Prefers terse openings.** Uses "btw", "actually", "oh
  and" to jump straight to the point. Evidence: 2026-04-24
  session (multiple turns), 2026-04-25 session (two turns).

Not:

## Communication style

- User is impatient. (No.)
- User likes short messages. (Too vague.)

Re-synthesise, don't blindly append

Every ~5 new observations (or at session close if a meaningful pattern emerged), re-read the whole file:

1. Merge duplicates that grew into redundant bullets.
2. Refine wording where sharper observations replaced earlier guesses.
3. Remove observations that turned out wrong or were one-offs — keep the profile refutable and honest.
4. Update last-synthesised: YYYY-MM-DD in the frontmatter.

The file is synthesis, not history. If you want the history, the session journals have it.

Section invariants

• Keep the "How to apply this profile" section at the bottom concrete and actionable. It's what other skills and agents actually consume. Bullets should be prescriptive: "Prefer X over Y," not observations.
• Keep "Open tensions" honest — if two observations contradict, note it there rather than picking a winner prematurely.
• Don't let any section grow past ~8-10 bullets. If it does, you're probably not re-synthesising; cluster or re-cut.

At session close

Invoked from cairn-close-session:

1. Read today's session journal.
2. Scan for signal-bearing exchanges (user correction, confirmed unusual choice, preference reveal).
3. If one or more new observations are warranted, add them with evidence.
4. Consider whether this triggers a re-synthesis pass (if >5 observations added since last synthesis, or if a newer observation supersedes an older one).
5. Log the update in the close-session handoff — one-liner saying "profile updated with N new observations" or "no new profile signal this session."

Privacy

The user profile is personal. Default location is private (~/.config/cairn/, not tracked in any repo). If the user chose a project-local tracked location at install time, they accepted the tradeoff — but the skill should still err toward private: don't save observations that would be embarrassing in a PR review.