Changelog

Changelog

v0.5.0

August 22, 2026

• Canonical agent directives are ~24% leaner at the source (~21% as shipped in generated plugins): the bounded-repair contract lives once in the global rules, specialist skill catalogs list only verified skills, and the delegation guard was re-anchored to the current canonical heading.
• Delegated implementation outcomes now complete at a pushed feature branch with an open PR; failed or cancelled delegations get one adjusted-brief retry before structured blocker reporting.
• Decision record: ADR-CORE-019 (docs/adr/core/ADR-CORE-019-directive-simplification.md).

v0.4.16

August 13, 2026

• Agent directives now calibrate effort to task risk, prefer mature ecosystem solutions, converge reviews on material blockers, deliver routine engineering work through feature branches and PRs, and clean up task-owned background processes before completion.

v0.4.15

August 12-13, 2026

• Clarified agent workflow contracts: routine validated commits on recognized feature branches remain autonomous after required review; push and later lifecycle actions stay separately gated.
• Simplified and re-aligned the shared directives around outcome, evidence, runtime authority, blind review, bounded repair, and autonomous routine work.

v0.4.14

August 11, 2026

• Streamlined canonical agent directives by centralizing universal contracts and adding bounded autonomy, work-unit budgets, scope control, process lifecycle evidence, and checkpoint action boundaries.

v0.4.13

August 6, 2026

• Delegation is more selective, handoffs stay concise, and Sonar research starts with the owning specialist, adding another specialist only when a distinct required output remains.

v0.4.12

August 3, 2026

Selective routing contract

The orchestrator now routes by task class across direct, focused, and full routes. Small explanations and tiny edits run directly or through one specialist; complex or high-risk work uses the full pipeline. fein selects the full route, sonar is research-only, and blitz is an explicit low-risk/direct bypass that does not waive safety floors.

v0.4.11

July 29, 2026

Rule-override conditions and communication conventions

• Added 6 explicit override conditions to the orchestrator prompt (user skip request, safety, mode override, frustration escalation, rule conflicts, explanation requests) so agents know when and how to deviate from defaults.
• Two new shared rules: report errors matter-of-factly (state problem, cause, fix without hedging) and lead with the action (first line actionable, context follows).
• Review triage now chains directly into the commit flow after approval, guarded against unresolved [fix]/[escalate] items.

v0.4.10

July 28, 2026

Blind review and fail-loud iteration exit

• The reviewer no longer receives the builder’s handoff notes or self-assessment; it evaluates only the diff, requirements, and acceptance criteria.
• When the review loop runs 3 cycles with unresolved issues, the pipeline stops and escalates with a structured report of what is still blocking, requiring an explicit override to continue.

v0.4.9

July 22, 2026

Directive recomposition

• Restructured core prompts with clearer sections and emphasis on critical rules.
• Added structured handoff verification checklists to all specialist agents.
• Standardized “before reporting done” completion checks across all agents.
• Added a parallelization table and multi-lens review swarm capability.
• Made prompt instructions platform-agnostic for consistent behavior across OpenCode, Cursor, Kimi Code, Pi, and other platforms.

v0.4.8

July 21, 2026

• Dependency updates across packages (diff, zod, @clack, effect, astro, and more). No user-facing behavior change.

v0.4.7

July 21, 2026

Switch to npm-based install

The plugin now installs via npm pack @maestria/kimi-code instead of git-based codeload. maestria update kimi-code no longer re-downloads the same version.

v0.4.6

July 21, 2026

Compact agent directives

Agent directives compressed to cut context usage by ~38% on the orchestrator prompt without removing rules or intent. Shared specialist rules now live once in global rules instead of being repeated per specialist.

v0.4.5

July 14, 2026

Structured assumption tagging

Adventurer and architect outputs now require [verified] or [inferred] tags on every assumption.

Deterministic agents, cognitive hygiene, outcome specs, and more

Additional methodology improvements including cognitive hygiene checklist, outcome over activity specs, experiment framing, parallel speculation, incidental findings, and first-principles decomposition.

v0.4.4

July 14, 2026

Structured PR descriptions

Pull request descriptions now follow a consistent format with Summary, Changes, Testing, and Breaking Changes sections. Reviews are faster when every PR uses the same structure.

v0.4.3

July 14, 2026

Auto-create feature branch when landing on main

The agent no longer asks for confirmation before creating a feature branch when it lands on main.

Work results format streamlined

The post-task summary now uses a table-first format. Changes are prefixed with + (added), ~ (modified), or - (deleted). Breaking changes are marked with !, and test file changes are annotated with (test).

v0.4.2

July 14, 2026

Structured work results are now mandatory

After every task, the agent must produce a ## Changes summary table. Documentation audit before every commit is also now required.

v0.4.1

July 10, 2026

• Reordered conventional commit prefixes. refactor is now the default prefix instead of feat. feat is reserved for user-facing features only.

v0.4.0

July 10, 2026

Multi-lens review swarm

The orchestrator can now dispatch parallel reviewers with different focus areas (Security, Architecture, Performance, UX, General). Issues are categorized [fix]/[dismiss]/[escalate].

Autonomous commit protocol

Commit is fully autonomous - agent reads git log, composes correct conventional commit message, commits without asking. Push is automatic on feature branches, asks only on main/master.

Eliminate questions - autonomous philosophy

Agents now exhaust data, document assumptions, and proceed instead of asking for permission. Mid-phase questions eliminated across all specialists.

Review methodology improvements

• Observation over reasoning - reviewer prioritizes running code over logical argument.
• Review triage pipeline - issues categorized and validated with conservative conflict resolution.

v0.3.5

July 1, 2026

Code intelligence tool preference

Agents now prefer code intelligence tools (MCP, CLI, or other) before falling back to grep and read when exploring the codebase.

v0.3.1

June 27, 2026

Two new design principles

Agents now check from first principles before adopting existing patterns, and check for existing solutions before building from scratch.

v0.3.0

June 24, 2026

Project workflow configuration (.maestria/)

Ships the .maestria/ project workflow protocol. Projects using maestria’s agent directives can now define .maestria/workflow.md for custom delegation sequencing and .maestria/rules.md for project-specific rules. The orchestrator loads these via @adventurer delegation.

Effort anthropomorphism guard and writing style guidance

The orchestrator and all specialists now include explicit guardrails against two common failure modes:

• Don’t anthropomorphize effort - A new Critical Rule tells the dispatcher that thinking “that analysis would be too much work” is a category error for a machine that only routes. The same principle is instilled in all specialists via the global agent rules.

v0.2.1

June 24, 2026

Single source of truth for agent directives

Agent prompts and global rules now live in one canonical location (packages/core/agent-directives/) and sync automatically to all platform plugins. The scripts/sync-all and scripts/check-sync tools handle the pipeline.

0.1.0

Minor Changes

• Initial release of @maestria/kimi-code - a declarative, manifest-based plugin that ports the 8 specialist skills from @maestria/opencode v0.3.2 to Kimi Code v0.13.1.
• 8 SKILL.md files - orchestrator + 7 specialists using the v0.3.x Skill Prescription pattern
• Manifest-only plugin - no SDK hooks, no build step; the plugin is pure static config (kimi.plugin.json + skills/ + rules/)
• Swarm-aware orchestration - the orchestrator routes ≥3 uniform items through Kimi Code’s AgentSwarm for parallel fan-out, with the 7-specialist table covering the coder / explore / plan subagent profiles
• Global rules - rules/AGENTS.md with the 7-specialist delegation table, auto-loaded by Kimi Code from ~/.kimi-code/AGENTS.md
• Install checklist - INSTALL.md with recommended [[hooks]] and [[permission.rules]] config blocks for tool-layer safety enforcement (persona text alone is advisory)
• 20 vitest tests - validate manifest schema, frontmatter, rules budget, and persona safety constraints (reviewer no-edit, adventurer read-only Bash)

See ADR-KC-001 for the full design rationale.