feat(interface): measure complete profile changes

[drewstone]

What changed

• let the existing AgentProfileDiff improvement contract accept every current AgentProfile axis
• map prompt, skills, tools, MCP, hooks, and subagents to their existing granular surface labels, with agent-profile covering residual axes
• require newly introduced measured resources to be inline so experiment and activation bytes stay exact
• reject structurally empty removal steps that cannot change a profile
• fix changed-axis detection for explicit empty identity values

No candidate type or parallel improvement record is added.

Proof

• pnpm test — all workspace tests passed; Agent Interface 228/228
• pnpm --filter @tangle-network/agent-interface check-types
• pnpm build
• git diff --check
• git merge-tree --write-tree origin/main HEAD

[tangletools]

✅ Auto-approved drewstone PR — 02f7ee07

This PR was opened by the trusted drewstone account.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: drewstone_author · 2026-07-29T21:08:00Z}

[tangletools]

🟢 Value Audit — sound

Verdict	sound
Concerns	0 (none)
Heuristic	0.0s
Duplication	0.0s
Interrogation	185.8s (2 bridge agents)
Total	185.8s

💰 Value — sound

Broadens the measured profile-improvement contract from prompt+skills-only to every AgentProfile axis, reusing the existing surface enum and axis-diff helper; coherent and in-grain, with two correct bug fixes bundled in.

• What it does: The agentProfileImprovementChangeStepSchema previously rejected any profile diff touching axes other than prompt and inline resources.skills. This change (1) removes that allow-list, so a measured/activated improvement may now set or remove model, harness, permissions, tools, mcp, connections, subagents, hooks, modes, confidential, metadata, extensions, and all resource kinds (files/tools/agen
• Goals it achieves: (1) Let the optimizer measure and ship improvements to ANY profile axis, not just prompt/skills — the prior restriction was self-described in code as 'the first product path', so this is the anticipated evolution. (2) Preserve byte-exact experiment/activation guarantees for ALL measured resources, closing a hole where a non-skill resource (e.g. a github-ref'd tool/file) could drift between signed
• Assessment: Sound and well-scoped. It extends existing primitives rather than forking them: the AgentImprovementSurface enum (agent-candidate.ts:818) already defined prompt/skills/tools/mcp/hooks/subagents/agent-profile, so the mapper consumes labels that already existed for other improvement types; changedAgentProfileAxes (profile-diff.ts:288) already enumerated every property axis exhaustively via agentProf
• Better / existing approach: none — this is the right approach. Searched for an existing axis→surface mapper (grep AgentImprovementSurface across packages/agent-interface/src: only agent-candidate.ts defines it and only this schema consumes it for profile improvements), an existing 'is this removal a no-op' predicate (grep hasRemovalOperation/isEmptyRemoval: none existed; this is the first), and any parallel profile-improveme
• Model: opencode/zai-coding-plan/glm-5.2
• Bridge attempts: 2
• Bridge warning: opencode/kimi-for-coding/k2p7: bridge stream ended without value-audit content

🎯 Usefulness — sound

Widens an existing, fully-wired profile-improvement contract to accept every AgentProfile axis (previously prompt+skills only), reusing the established surface vocabulary and inline-bytes rule with no new types or competing patterns.

• Integration: Live and reachable. agentProfileImprovementChangeSchema (agent-profile-improvement-schema.ts:200) feeds agentProfileImprovementExperimentSchema (line 213) → agentProfileImprovementMeasuredComparisonSchema (line 462) → the evaluation discriminant of agentImprovementProposalSchema (agent-candidate-promotion-schema.ts:449). The rewritten changedProfileImprovementSurfaces is called at agen
• Fit with existing patterns: Follows the codebase grain precisely. Reuses the existing AgentProfileDiff/agentProfileDiffSchema as the patch base rather than inventing a new format; reuses the existing AgentImprovementSurface enum from agent-candidate.ts:818 (no new labels — prompt/skills/tools/mcp/hooks/subagents already existed, agent-profile was already the catchall); reuses changedAgentProfileAxes from profile-di
• Real-world viability: Edge-case handling is careful and tested. Set-side resources map granular keys (skills/tools/agents→subagents) vs. residual→agent-profile (lines 518-525); remove-side distinguishes resources: true (full removal → all granular + catchall, lines 529-533) from partial removals, filtering empty arrays so structurally empty removals don't emit false surfaces (lines 534-543). The empty-{}-resources
• Model: opencode/zai-coding-plan/glm-5.2
• Bridge attempts: 1

No concerns — sound change, no better or existing approach found. ✅

---

What this audit checks

It judges the change on its merits — not whether it was tasked out in an issue. Unticketed, fast-moving work is fine; the question is whether the change is good and whether a better or existing approach should be used instead.

Pass	What it asks
Heuristic	Vague title? Whitespace-only or cruft-bearing diff? (content signals only)
Duplication	Do added function/class names already exist elsewhere in the repo?
Value Audit	What does it do? What goal does it achieve? Is it good? Better architecture or already-exists?
Usefulness Audit	Does it integrate and fit? Will it hold up in real use and actually get used?

Findings are concerns, not blocks — the human reviewer decides what to do with them.

_{value-audit · 20260729T211127Z}