Open
Conversation
Complete opencode configuration including: - Agent definitions (12 agents: senior-engineer, qa-engineer, tech-lead, etc.) - Slash commands (56 commands: /commit, /pr, /test, /review, etc.) - Skills library (159 skills covering development, testing, architecture, communication) - Core configuration files (opencode.json, optimized config) This establishes the full opencode agent system with progressive skill loading, mandatory always-active skills (pre-action, memory-keeper, token-efficiency), and comprehensive domain expertise for software engineering workflows. Files: 196 changed (+8200, -743)
There was a problem hiding this comment.
Pull request overview
This PR establishes a complete opencode agent configuration system for AI-assisted software development workflows. The implementation includes 12 agent definitions, 56 slash commands, configuration files, and claims 159 skills. However, analysis reveals that the majority of skill files (~100+) contain only placeholder template text rather than actual implementations.
Changes:
- Added opencode configuration files with MCP server setup and Ollama provider configuration
- Created 12 agent definitions (senior-engineer, qa-engineer, tech-lead, devops, security-engineer, data-analyst, writer, sysop, linux-expert, nix-expert, embedded-engineer)
- Created 56 command definitions for development workflows
- Added ~159 skill files, though most contain only placeholder template content
- Established core AGENTS.md documentation defining mandatory requirements and workflows
Reviewed changes
Copilot reviewed 196 out of 196 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| opencode.json | MCP server configuration and plugin setup |
| opencode-local-optimized.json | Ollama local provider configuration with model definitions |
| AGENTS.md | Core documentation defining mandatory requirements and orchestration rules |
| agents/*.md | 12 agent definitions with roles and skill loading patterns |
| commands/*.md | 56 command definitions for workflows (commit, test, review, etc.) |
| skills/*.md | 159 skill files - majority contain placeholder templates only |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+16
to
+34
| ## Core principles | ||
|
|
||
| 1. Principle one | ||
| 2. Principle two | ||
| 3. Principle three | ||
|
|
||
| ## Patterns & examples | ||
|
|
||
| Include concrete examples relevant to this skill. | ||
|
|
||
| ## Anti-patterns to avoid | ||
|
|
||
| - ❌ Common mistake one | ||
| - ❌ Common mistake two | ||
|
|
||
| ## Related skills | ||
|
|
||
| - `skill-a` - Pairs with this skill | ||
| - `skill-b` - Alternative approach |
There was a problem hiding this comment.
This skill file contains only placeholder template text and has not been properly implemented. The "Core principles" section lists generic "Principle one", "Principle two", etc. instead of actual principles. The "Patterns & examples" section says "Include concrete examples" without providing any, and the "Related skills" section references non-existent "skill-a" and "skill-b" placeholders.
This pattern appears across approximately 100+ skill files in this PR (based on search results showing "Principle one" in 100+ files). These placeholder skills provide no actual value and may cause confusion when loaded during agent execution.
All skill files should either be:
- Fully implemented with real content, OR
- Removed until they can be properly implemented
The PR description claims "159 Skills" with "Comprehensive coverage" but the reality is most skills are empty templates.
- Add BATS testing framework for skill import pipeline - Create test helper with setup/teardown utilities - Add skill-discovery skill for proactive skills.sh suggestions - Create vendor directory structure for external skills - Initialize skill lockfile for version tracking AI-Generated-By: Opencode (kimi-k2.5-free) Reviewed-By: Yomi Colledge
- 13 tests covering import, remove, and collision detection - Mock git repos for isolated testing (no network required) - Test isolation with temp directories (no ~/.config/opencode changes) - Coverage: skill-import, skill-remove, collision detection, error handling - All tests pass in ~6.5 seconds AI-Generated-By: Opencode (kimi-k2.5-free) Reviewed-By: Yomi Colledge
Updates AGENTS.md with comprehensive agent system documentation including: - Three Pillar Model (always-active discipline, parallel execution, progressive disclosure) - Commit Rules with hybrid workflow (git_master planning + make ai-commit execution) - Model Routing Strategy with three-tier system (T1/T2/T3) - Change Request Verification workflow with evidence requirements - Mandatory skills and category system This is the foundational documentation that defines all agent behaviors.
- Adds senior-engineer.md: New agent definition for senior engineering tasks - Updates commit.md: Documents the hybrid ai-commit workflow Implements agent system with proper command definitions.
…kflow Adds three new comprehensive skills: - code-reading/SKILL.md: Codebase navigation and understanding guidance - debug-test/SKILL.md: Test debugging workflow patterns - tdd-workflow/SKILL.md: Red-Green-Refactor cycle implementation Expands skill library with essential development workflows.
Updates foundational skills with improved documentation: - architecture: Enhanced architectural pattern guidance - api-design: RESTful conventions and versioning - clean-code: SOLID principles and Boy Scout Rule - code-reviewer: Comprehensive review checklist - code-generation: go:generate and template patterns - design-patterns: Pattern recognition and application Improves code structure and quality guidance.
Enhances testing, error handling, and validation documentation: - error-handling: Language-agnostic patterns and strategies - refactor: Systematic refactoring with safety nets - prove-correctness: Writing tests to verify behavior - bdd-workflow: Behavior-Driven Development guidance - cucumber/ginkgo-gomega/godog: BDD framework implementations Improves test coverage and code reliability guidance.
Updates language-specific testing framework documentation: - jest: Jest testing framework patterns - cypress: E2E testing for web applications - rspec-testing: RSpec BDD for Ruby - bubble-tea-testing: TUI testing with Bubble Tea - huh-testing: Form component testing - e2e-testing: End-to-end testing patterns - fuzz-testing: Fuzzing for edge cases Expands testing coverage across languages and frameworks.
Enhances Go language and concurrency expertise: - golang: Idioms, patterns, and best practices - concurrency: Goroutines, channels, and sync primitives - performance: Optimization and profiling - gomock: Mock implementation generation - gorm-repository: GORM ORM and repository patterns Improves Go development guidance with comprehensive patterns.
Updates terminal UI and scaffolding documentation: - bubble-tea-expert: Bubble Tea TUI framework expertise - huh: Interactive form library patterns - create-screen: Screen component creation - create-bug: Bug report scaffolding - create-intent: Intent creation with architecture - create-pr: Pull request creation workflow - create-task: Task creation with acceptance criteria Enhances TUI development and artifact creation workflows.
Enhances API design and validation workflows: - graphql: GraphQL API patterns - db-operations: Repository patterns with GORM/SQLite - pre-action: Pre-execution decision framework - pre-merge: Final validation checklist - check-compliance: Full compliance verification Improves API development and quality gates.
Enhances content creation and documentation workflows: - documentation-writing: Technical documentation patterns - api-documentation: API documentation best practices - accessibility-writing: Accessible content guidelines - blog-writing: Technical blog post writing - presentation-writing: Conference and talk preparation - tutorial-writing: Step-by-step learning guides - proof-reader: Editing and clarity verification Improves documentation quality and communication.
Enhances Git operations and repository management: - git-advanced: Rebasing, cherry-picking, and history management - git-worktree: Parallel development with worktrees - auto-rebase: Automatic PR rebasing and conflict resolution Improves Git workflows and collaboration patterns.
Enhances production operations and monitoring: - logging-observability: Structured logging and tracing - monitoring: System health checks and observability - devops: CI/CD and infrastructure as code Improves production reliability and operational excellence.
Enhances reasoning, analysis, and decision-making: - critical-thinking: Rigorous analysis and validation - assumption-tracker: Explicit assumption tracking - devils-advocate: Stress-testing solutions - question-resolver: Systematic problem resolution - justify-decision: Evidence-based decision justification - trade-off-analysis: Systematic trade-off evaluation - systems-thinker: Complex system understanding - epistemic-rigor: Intellectual honesty frameworks Improves analytical capabilities and decision quality.
Enhances task execution and efficiency optimization: - checklist-discipline: Rigorous checklist discipline - task-completer: Comprehensive task completion - task-tracker: Progress tracking with scoring - time-management: Timeboxing and focus techniques - parallel-execution: Running independent tasks - scope-management: Scope control and creep prevention - token-efficiency: AI interaction value optimization - token-cost-estimation: Work estimation and tracking Improves workflow efficiency and resource management.
Enhances cloud, deployment, and hardware development: - aws: AWS cloud services (EC2, ECS, S3, Lambda, RDS) - bare-metal: Physical server provisioning - heroku: PaaS deployment patterns - virtual: VPS and virtualization hosting - nix: Nix package manager and flakes - cpp: C++ for embedded systems - platformio: PlatformIO build system - embedded-testing: Hardware-in-the-loop patterns Expands infrastructure and embedded development capabilities.
Enhances security and operational incident handling: - cyber-security: Vulnerability assessment and defense - security: Secure coding practices - incident-response: Production incident handling - incident-communication: Professional incident communication Improves security posture and incident management.
Enhances cross-cutting utility capabilities: - accessibility: Terminal application accessibility - ai-commit: Properly attributed AI commit workflow - automation: CI/CD pipelines and self-maintaining systems - benchmarking: Go performance benchmarking - british-english: British spelling conventions - configuration-management: Environment and secrets handling - dependency-management: Go modules and versioning - breaking-changes: Backwards compatibility strategies Improves utility and support capabilities.
Enhances knowledge sharing and release workflows: - knowledge-base: Knowledge graph management - memory-keeper: Discovery and solution capture - mentoring: Junior engineer guidance - research: Systematic investigation - investigation: Codebase analysis documentation - release-management: Versioning and changelog management - release-notes: Clear release documentation - rollback-recovery: Failed deployment handling Improves knowledge retention and release processes.
Enhances domain modeling and communication: - migration-strategies: Safe schema and data migrations - domain-modeling: DDD and domain patterns - email-communication: Professional technical emails - feature-flags: Safe feature rollouts - fix-architecture: Architecture violation diagnosis - information-architecture: Content structuring - mongoid: MongoDB ORM patterns Improves domain expertise and communication.
Enhances PR monitoring and development workflows: - monitoring: Post-deployment observability - new-skill: Skill creation with integration - note-taking: Obsidian and documentation notes - pair-programming: Driver/navigator collaboration - pr-monitor: PR status and review coordination - profiling: Performance bottleneck identification - core-auto-detect: Environment detection Improves development workflows and PR management.
Enhances Obsidian knowledge management capabilities: - obsidian-chartjs-expert: Chart embedding - obsidian-codeblock-expert: Syntax highlighting - obsidian-consolidation: Zettelkasten refinement - obsidian-customjs-expert: Custom scripting - obsidian-dataview-expert: Dynamic queries - obsidian-frontmatter: Metadata management - obsidian-latex-expert: Mathematical notation - obsidian-mermaid-expert: Flowcharts and diagrams - obsidian-structure: PARA structure enforcement Improves Obsidian vault management and productivity.
Enhances language-specific and review capabilities: - estimation: Work estimation and uncertainty - github-expert: GitHub Actions and workflows - javascript: JS/TS and Vue.js patterns - pragmatic-problem-solving: Practical solution focus - respond-to-review: Review feedback responses - retrofitting-types: Gradual type addition - retrospective: Post-mortem and learning - ruby: Ruby/Rails development patterns Expands language expertise and collaboration.
Enhances backend and quality assurance capabilities: - scripter: Bash and Python automation - service-layer: Business logic orchestration - sql: Query optimization patterns - static-analysis: Code analysis tools - style-guide: Documentation conventions - tool-usage-discipline: MCP and skills usage Improves service layer and code quality practices.
Add MODEL: log entry to failover.log showing which provider/model each agent session uses. Only fires once per session or when the model changes, avoiding noise from repeated chat.params hook calls. Remove unused newHealthKey variable in the proactive switch block.
The chat.params hook was proactively switching ALL sessions including orchestrators (atlas/sisyphus), causing the parent session to lose its model. Now: - Agent name → tier mapping routes subagents to correct tier models - Orchestrator sessions are never proactively switched - Model and provider are always set as a pair (fixes big-pickle/anthropic bug) - Unknown agents default to T2; unknown/missing agent name = orchestrator
The chat.params hook was proactively switching ALL sessions including orchestrators (atlas/sisyphus), causing the parent session to lose its model. Now: - Agent name → tier mapping routes subagents to correct tier models - Orchestrator sessions are never proactively switched - Model and provider are always set as a pair (fixes big-pickle/anthropic bug) - Unknown agents default to T2; unknown/missing agent name = orchestrator
HealthManager.clearExpired() removed stale rate-limit entries from memory but never wrote back to disk. Stale entries persisted in provider-health.json between sessions until the next flush() call. Now atomicWriteSync() is called immediately after clearExpired() in the constructor, ensuring the JSON file is clean on disk from the moment the plugin loads.
Close the loophole where orchestrators were classifying investigation as "coordination" to justify direct bash/read/glob/grep usage. Add explicit tool usage constraints to AGENTS.md and enforce via prompt restrictions in oh-my-opencode.jsonc. All investigation work must be delegated to explore or Researcher agents. Only binary verification (build status, test results, git status) is allowed directly. Final read-back to confirm completed work is the sole exception.
…ast notifications - Add anthropic/claude-haiku-4-5 to T1 chain (after copilot haiku) - Anthropic sonnet models already in T2 chain (lines 77-78) - Anthropic opus model already in T3 chain (line 98) - Remove misplaced big-pickle entry from T3 array (was T2 tier) - Make toast notifications conditional on actual rate limits (not preference switches) - Prevents noisy notifications when fallback chain prefers different model than parent session
…inking options on switch - Proactive routing now checks isRateLimited(healthKey) before switching — healthy models are never replaced - Add clearThinkingOptions() to remove thinking/effort/thinkingConfig keys when switching to non-Claude models - Add modelSupportsThinking() to detect Claude Opus/Sonnet models that support extended thinking - Rename _output to output in chat.params hook so thinking options can be written - Prevents invalid API params when falling back from a Claude model to GPT/Gemini/big-pickle
Add Knowledge Lookup Protocol, KB Curator Auto-Trigger Protocol, and Skill Injection Limits sections to AGENTS.md. Update Three Pillars to Knowledge-First. Add PARALLEL EXECUTION mandate to orchestrator prompt_append blocks. Replace verbose MANDATORY DISCIPLINE with compact KNOWLEDGE LOOKUP block across all agents. Silence agent routing chatter.
Pretty-print single-line JSON blob with 2-space indentation for human readability. No content changes.
Structured workflow for handling PR review feedback systematically, covering evaluation, classification, implementation and evidence reporting.
Add keyword pattern for PR review feedback terms so the auto-loader injects the pr-review-workflow skill when relevant tasks are detected.
…efault skills Ensures the Code-Reviewer agent has the complete PR lifecycle covered: triage (pr-review-workflow), implementation (respond-to-review), and final validation (pre-merge).
The bare 'claude-sonnet-4' identifier was missing the '-0' version suffix, causing incorrect model references in the provider failover chain.
Enforces intelligent agent selection over Sisyphus Junior catch-all, single-concern task atomicity, 15-task session cap, and emergency-only exceptions. Applies to all orchestrators.
Adds priority-based, case-insensitive regex matching against agent_patterns config. Returns highest-priority match or null result when no patterns match or prompt is empty. Also adds agent_patterns field to SkillAutoLoaderConfig interface.
Add TEST-FIRST MANDATE (NON-NEGOTIABLE) block to the prompt_append of Senior-Engineer, QA-Engineer, Embedded-Engineer, Tech-Lead, and sisyphus-junior. Mandates Red-Green-Refactor cycle before any implementation code is written. Non-implementation agents excluded.
…reset Update GitHub Copilot monthly threshold to 300 requests. Implement calendar-month-based reset for monthly periods instead of rolling intervals. Improve provider extraction to distinguish between claude and anthropic models with explicit provider ID handling.
…mplementary phases pre-action reframed as PREFLIGHT schema with role-specific fields (orchestrator/worker/read-only). parallel-execution reframed as EXECUTE phase that runs after Preflight planning. Both cross-reference each other. Resolves contradictory instructions where one said 'stop and think' while the other said 'batch immediately'.
…finitions Adds mandatory Step Discipline block after frontmatter in every agent .md file. Enforces permission chain (User → Orchestrator → Sub-agent), prohibits self-authorised step skipping, and defines what counts as skipping (omitting steps, placeholders, nolint markers). Previously only Knowledge Base Curator.md had this — now all 17 agents do.
Rewrites all 24 agent prompt_append strings from 30KB+ bloated prompts to role-specific CRITICAL blocks (546-1242 chars each). Five templates: orchestrator, worker, writer, read-only, lookup — each with mandatory PREFLIGHT schema. File size reduced 23% (34KB → 27KB). Includes reusable Python script at scripts/rewrite-prompt-append.py for future prompt_append changes (supports --dry-run, --backup).
…rnance, and model diversity - Rewrite all 28 agent prompt_appends to lean versions with mcp_skill() loading - Restrict agent-discovery to orchestrators only (sisyphus, hephaestus, atlas, Tech-Lead) - Add 6 thinking skills to Prometheus (built-in OMO agent, not custom file) - Workers use ralph loop completion signal (<promise>DONE</promise>) - Add category model overrides: deep/ultrabrain → gpt-5.2-codex, visual/artistry → gemini-3-pro - Reorder failover chains for model diversity (GPT-5 leads T2, GPT-5.2 leads T3) - Disable skill content injection, replace with lean skill name line in prompt - Remove dead code: skill-content-injection.ts, mcp-mem0-server-sdk.ts, old agents-rules files - Add agentic health check script (make health-check, 5/5 passing) - Fix stale tests in skill-selector, agent-config-parser, orchestrator-only test files - Add compliance specs, discipline skill, and orchestrator compliance tooling - All 301 tests passing, health check 5/5
Replace vague "after significant work" KB Curator trigger with explicit categories: (1) project/feature work, (2) exploration/investigation, (3) agentic flow/config changes. Updated in orchestrator prompt_appends, discipline skill, and AGENTS.md. Marked as non-negotiable blocking violation.
Allow all agents to work in git worktrees outside the main working directory by setting external_directory to allow. Add safety rules protecting main and next worktrees from modification without explicit user permission. Rules enforced via AGENTS.md and discipline skill.
Signed-off-by: Yomi Colledge <baphled@boodah.net>
…l execution guidance Fix rate-limited providers being selected by subagents due to key mismatch between fallback chain keys and inferred provider keys. - Add isModelRateLimitedByAnyProvider() to HealthManager for cross-provider detection - Restore if(isRateLimited) guard in chat.params accidentally removed - Expand rate limit keywords: free usage exceeded, exceeded, add credits, quota - Remove unavailable opencode/gpt-5-nano from T1, fix duplicate chain entries - Add 4 new recommend-model tests (15/15 passing) - Add parallel-execution to universal auto-load skills in AGENTS.md - Strengthen parallel guidance in all 18 worker/specialist prompt_appends
- Delete opencode-local-optimized.json (deprecated configuration) - Update PR-related skills: create-pr, pr-monitor, pr-review-workflow, respond-to-review - Update VHS skill with expanded guidance - Minor updates to bdd-workflow, bubble-tea-expert, ui-design skills
baphled
force-pushed
the
opencode-config-complete
branch
from
115f856 to
9ca108c
Compare
baphled
added a commit
to baphled/dotopencode
that referenced
this pull request
Mar 5, 2026
Complete OpenCode configuration including: - Agent definitions (17 agents: Senior-Engineer, QA-Engineer, Tech-Lead, etc.) - Slash commands (47 commands: /commit, /pr, /test, /review, etc.) - Skills library (154 skills covering development, testing, architecture, communication) - Core configuration files (opencode.json, oh-my-opencode.jsonc) - Plugin system with extensions - Test suite and automation scripts This establishes the full OpenCode agent system with: - Progressive skill loading - Mandatory always-active skills (pre-action, memory-keeper, discipline) - Three-tier model routing (T1/T2/T3) - Agent orchestration and delegation patterns Extracted from: baphled/dotfiles#5
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR establishes the complete opencode agent system configuration for AI-assisted software development workflows.
Changes include:
Key Features:
Impact:
Files Changed: 196 files (+8,200 insertions, -743 deletions)
AI Attribution: This work was created with assistance from Claude Code (claude-sonnet-4-5)