feat(persona): graph-capability awareness, read-only boundary kept — 0.22.5

[Cain-Ish]

Why

The persona-as-collaborator protocol (using-second-brain) knew the graph read tools + the wingman graph-check, but never referenced knowledge_relate — so the persona-driven loop never closed the loop on a relationship confirmed mid-session. Goal: persona aware of the capability, graph accrues with no user interaction.

What

The wingman now surfaces a suggested knowledge_relate when the work confirms (or retires) a typed relationship — confirmed/retired only, never speculative. It does not call the tool: edge curation stays owned by the three sanctioned writers (capture-time extractor + the user's manual tool + knowledge-maintainer), and the extractor records the relationship from the session transcript at Stop. So the graph still accrues without user interaction, and the persona stays a reader.

Design note — release-gate review drove this

An earlier cut granted knowledge_relate in the persona's allowed-tools (direct write). The history/regression reviewer flagged it HIGH: that makes the always-on, model-invoked persona a 4th, ungoverned graph writer, reverting the deliberate write-ownership boundary agents/dream-runner.md enumerates, and pushing drain load onto the maintainer for edges the persona was never meant to make. The architectural reviewer independently flagged the "confirmed" trigger as too fuzzy. Per those findings (and an explicit maintainer call), the persona surfaces, it does not write — keeping the boundary intact while still delivering the awareness + zero-interaction accrual.

Guard

New tests/test-persona-capability-awareness.sh pins both halves: awareness of knowledge_relate AND the read-only allowed-tools boundary (the tool must NOT be granted). Incorporates the unit reviewer's robustness fixes (.* over [^\n]*; asserts the load-bearing "never speculative" clause).

Verification

Full suite green (59 shell + 265 vitest, 0 fail); version lockstep 0.22.5; validate-plugin OK. Prompt/test-only — no MCP change, no state migration.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Adds awareness of knowledge_relate to the persona-as-collaborator protocol while explicitly preserving the read-only boundary (the persona surfaces a suggestion; capture-time extractor + maintainer remain the writers). Bumps version to 0.22.5 and adds a guard test pinning both halves (awareness AND read-only allowed-tools).

Changes:

• New paragraph in skills/using-second-brain/SKILL.md instructing the wingman to surface a suggested knowledge_relate for confirmed/retired typed relationships only.
• New tests/test-persona-capability-awareness.sh asserting read tools granted, knowledge_relate NOT in allowed-tools, and the "never speculative" clause is present.
• Version bump (plugin.json, marketplace.json) and 0.22.5 entry in the upgrade migrations table.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
skills/using-second-brain/SKILL.md	Adds the surface-only knowledge_relate paragraph.
tests/test-persona-capability-awareness.sh	New guard test for awareness + read-only boundary.
skills/upgrade/SKILL.md	Adds 0.22.5 migration row.
.claude-plugin/plugin.json	Version → 0.22.5.
.claude-plugin/marketplace.json	Version → 0.22.5.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.