Event 71 · privatize 4 operator-profile canonicals + clarify skills/private/ semantic

[junjslee]

Summary

Closes the privatization audit started this session. After Events 65/66 privatized 14 docs/*.md files for operational state + positioning narrative, the only remaining operator-lived content surface was core/memory/global/{operator_profile, cognitive_profile, workflow_policy, agent_feedback}.md — the maintainer's actual filled-in profile data (cognitive axes, noise signatures, expertise scores, agent-learned behavioral rules). Privatizing these with the same symlink mechanism + clarifying the canonical-vs-example pattern in AGENTS.md governance.

Plus: clarify the skills/private/ naming. Pre-flight grep found three tracked files that document skills/private/ as a sync-exemption boundary (skills there are tracked publicly but NOT propagated by episteme sync), not as an operator-private content path. Renaming → skills/personal/ would have broken that contract; instead this commit clarifies the semantic in-place (Option B per operator authorization).

Pre-flight discipline note

Operator initially called the skills/private/ rename "just a naming quirk" and authorized → skills/personal/. Pre-flight git grep surfaced 3 tracked cross-refs documenting skills/private/ as a load-bearing sync-exemption boundary. Surfaced finding to operator → revised authorization to Option B (clarify-in-place, don't rename). The kernel's loss-averse posture working as designed; the asymmetry_posture audit is now firing in both directions across recent Events.

What this commit does

1. Privatize 4 operator-profile canonicals (Events-65/66 mechanism)

File	Action
core/memory/global/operator_profile.md	cp → ~/episteme-private/..., git rm, symlink
core/memory/global/cognitive_profile.md	same
core/memory/global/workflow_policy.md	same
core/memory/global/agent_feedback.md	same

Symlinks resolve transparently — verified via head -1 on each path returns the expected real-profile heading (# Operator Profile, # Cognitive Profile, etc.).

2. .gitignore — 4 new entries

Under a new "Privatized operator-profile canonicals (Event 71)" section. Symlinks at canonical paths float untracked; public commit captures pure deletion.

3. AGENTS.md — Doc classification policy extension

New subsection "Canonical-vs-example pattern (operator profiles in core/memory/global/)" under the existing classification policy. Documents:

• Privatize-via-symlink mechanism matches docs/*.md privatization
• examples/<name>.example.md ships PUBLIC as the fork-onboarding starting template
• Operator's lived state at ~/episteme-private/; symlinks resolve transparently
• Why operator's REAL profile is NOT shipped publicly as an "example" (identity-specificity vs sanitized template)
• Special case: agent_feedback.md has no .example.md template (purely operator-lived; forks initialize empty + accumulate)

4. skills/private/ semantic clarification (Option B)

Three files updated to clarify that skills/private/ is a sync-exemption boundary (skills tracked publicly but not propagated by episteme sync), distinct from operator-private content (which uses the ~/episteme-private/ symlink pattern):

• AGENTS.md L130 (Safe-to-edit-freely list)
• docs/CUSTOMIZATION.md L31
• docs/SKILLS_AND_PERSONAS.md L17

skills/private/skill-lab-autoresearch/ directory NOT renamed. The "private" name is load-bearing.

Operator session experience: UNCHANGED

Symlinks at canonical paths resolve to private staging. episteme sync + ~/.claude/CLAUDE.md @-imports continue reading the operator's real profiles transparently. The Event 68 audit-drift re-elicit (asymmetry_posture evidence_refs) remains intact at the private location.

Fork user experience: BETTER

Fresh clones see no canonical files at the privatized paths. episteme init seeds from examples/ — generic, identity-neutral, structurally-realistic templates. No risk of accidentally importing the maintainer's cognitive identity by copying canonical content.

Files in scope (8)

M  .gitignore                              ← append 4 Event 71 entries
M  AGENTS.md                               ← skills/private semantic + new
                                              Canonical-vs-example subsection
M  docs/CUSTOMIZATION.md                   ← skills/private semantic
M  docs/SKILLS_AND_PERSONAS.md             ← skills/private semantic
D  core/memory/global/operator_profile.md  ← DELETED (now symlink, gitignored)
D  core/memory/global/cognitive_profile.md ← DELETED (now symlink, gitignored)
D  core/memory/global/workflow_policy.md   ← DELETED (now symlink, gitignored)
D  core/memory/global/agent_feedback.md    ← DELETED (now symlink, gitignored)

Soak invariants

ZERO kernel/* / core/hooks/* / core/blueprints/* / src/episteme/* / tests/* / templates/* / labs/* touches. License unchanged. Plugin manifests unchanged. Kernel pre-commit chain (reviewer + test-runner agents) validated.

Held to Day-7+ (still pending)

• Day-7 GA-tag cut (gated on soak completion ~2026-04-30)
• gh-API signal-check trend re-run (~Day 6.5)
• git filter-repo decision (baseline NEGATIVE — likely unnecessary)

Deferred discoveries (logged for v1.0.1, not blocking)

1. scripts/restore-private-symlinks.sh — now covers 4 + 10 + 4 = 18 privatized symlinks. Operator-driven post-soak. Increasing utility with each Event that adds symlinks.

2. Verify episteme init seeds canonical paths from examples/*.example.md — this Event ASSUMES init does this; if not, fork onboarding for these 4 paths breaks until the CLI is patched. Verify post-merge by running episteme init in a fresh project (or read src/episteme/cli.py _init_memory directly). If missing, log as v1.0.1 CLI fix CP candidate.

3. agent_feedback template gap — no examples/agent_feedback.example.md exists. Acceptable per the Canonical-vs-example subsection rationale (agent-learned rules are operator-lived; forks initialize empty). If forks find this confusing, add a stub template in v1.0.1.

Test plan

• git diff master..HEAD --name-only matches exactly 8 files (no kernel/hot-path)
• head -1 core/memory/global/operator_profile.md returns # Operator Profile (symlink resolves to operator's real profile)
• Same for cognitive_profile / workflow_policy / agent_feedback
• git status --ignored shows the 4 canonical paths under "ignored"
• cat .gitignore | grep -A 5 'Event 71' shows the 4 new entries with the explanatory comment block
• AGENTS.md "Doc classification policy" section now contains "Canonical-vs-example pattern" subsection
• skills/private/skill-lab-autoresearch/SKILL.md still exists (not renamed)
• After merge: fresh clone of master + episteme init → check if core/memory/global/{operator,cognitive,workflow,agent}*.md get seeded from examples/ (deferred discovery chore(master): release episteme 1.1.0-rc1 #2)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
episteme	Ready	Preview, Comment	Apr 27, 2026 4:39pm