47e7087cc8dc8d5ffac9599d5c8838219c39f6b8
4
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Evan
|
d43077afa9 |
fix(providers): remove ~anthropic, skip ~ prefixes in sync script (#69)
* fix(providers): remove ~anthropic, skip ~ prefixes in sync script
OpenRouter uses ~ prefixes for internal auto-routing aliases (e.g. ~anthropic).
These are not real providers — they already route through openrouter.toml.
The generated ~anthropic.toml was confusing (looked like a stale backup)
and redundant with the existing openrouter provider.
- Delete providers/~anthropic.toml
- Skip provider IDs starting with ~ in sync-pricing.py --create-missing
* fix(providers): remove morph, aider, kwaipilot
- morph: specialized code-editing/patching tool, not a general LLM provider
- aider: CLI meta-tool wrapper (base_url empty), redundant with claude-code/codex-cli/gemini-cli/qwen-code
- kwaipilot: Kwai internal coding assistant routed via OpenRouter, niche
* fix(sync): add morph/aider/kwaipilot to SKIP_PROVIDERS to prevent re-creation
* feat(sync): merge OpenRouter-only providers into openrouter.toml
Instead of generating standalone .toml files that just wrap the OpenRouter
endpoint, merge their models directly into openrouter.toml with the
standard 'openrouter/{provider}/{model}' ID convention.
- Add _build_model_fields() and _model_lines() helpers to deduplicate
model rendering between standalone and merged paths
- Add merge_into_openrouter() that appends new models idempotently
- generate_provider_toml() now only runs for providers in PROVIDER_API
- --create-missing routes OpenRouter-only providers to merge_into_openrouter
* fix(providers): remove 14 OpenRouter-only standalone files
These providers have no direct public API and all route through
openrouter.ai/api/v1. Per the new sync-pricing.py policy, their models
will be merged into openrouter.toml on the next CI run instead of
living in separate files that just wrap the OpenRouter endpoint.
Removed: allenai, deepcogito, essentialai, inclusionai, inflection,
liquid, meituan, nex-agi, nousresearch, prime-intellect, relace,
switchpoint, tngtech, writer
* fix(providers): remove 7 niche providers with no driver support
No dedicated LLM driver code exists for these providers — they rely
purely on OpenAI-compatible passthrough with no special handling.
Removing them reduces registry noise; users can still reach them via
openrouter.toml if needed.
Removed: microsoft, ibm-granite, xiaomi, upstage, inception, aion-labs, arcee-ai
* fix(providers): remove ai21, chutes, venice
All three use ApiFormat::OpenAI with no special handling — pure passthrough.
No registry entry needed; users can reach them via openrouter.toml or by
adding a custom provider.
* docs(providers): rewrite README with full provider catalog and inclusion criteria
- List all 46 providers grouped by category with descriptions
- Document why each provider exists (direct API, unique endpoint, dedicated driver, local, CLI)
- Add inclusion criteria section explaining when to create standalone files vs merging into openrouter.toml
- Document sync script routing logic
- Update model counts: 49→46 providers, 339→232 models
* docs: add comprehensive READMEs for all registry sections + deepinfra provider
- agents/README.md: 32 agents across 7 categories with capability field reference
- channels/README.md: 44 channels across 5 categories with protocol reference table
- hands/README.md: 18 hands across 5 categories with HAND.toml format guide
- mcp/README.md: 33 MCP servers across 5 categories with transport/auth format
- plugins/README.md: 12 plugins with hook protocol documentation
- skills/README.md: 60 skills across 9 categories with SKILL.md format guide
- providers/deepinfra.toml: add DeepInfra serverless inference (5 models)
|
||
|
Evan
|
2b8259a0c0 |
feat(hands): add devteam hand (#41)
* feat(hands): add devteam hand -- autonomous software development team Multi-agent hand with 7 roles (PM, Architect, Frontend, Backend, DevOps, QA, Designer) and 3 team size tiers (simple/standard/full) for different project scales. PM coordinator auto-scans GitHub issues, triages, assigns tasks to specialists, and tracks progress on an in-memory project board. * refactor(hands): slim devteam to 3 agents (PM + Engineer + QA) 7 agents with serial agent_send = massive token waste and info loss at every handoff. Merge architect/frontend/backend/devops into one Engineer with full context. Keep QA separate for independent verification. Drop designer. Tiers: lite (PM + Engineer) and standard (PM + Engineer + QA). * fix(hands/devteam): fix workspace isolation and git workflow gaps - PM uses GitHub API for code browsing, no repo clone needed - Engineer explicitly clones repo, branches, commits, pushes, creates PR - QA explicitly clones repo, checks out branch under review - PM tracks last_scan timestamp to filter already-triaged issues - approval_mode now means PR stays open for review, not skip commit * fix(hands/devteam): use shared repo checkout instead of per-agent clones All 3 agents share one checkout at ../shared/repo/. Engineer clones it on the first task; PM and QA read from the same path. Eliminates duplicate clones and cross-workspace visibility issues. * fix(hands/devteam): read issue comments before triaging Comments contain clarifications, reproduction steps, duplicate markers, and resolution status. Also skip already-assigned and wontfix issues. * fix(hands/devteam): fix interactive git add, add merge/close APIs, add fix iteration flow - Replace git add -p (interactive) with git add <specific files> - PM prompt now has explicit merge PR and close issue API calls - Engineer has explicit fix-request handling (same branch, push, no new PR) * feat(hands/devteam): add full GitHub interaction -- PR review, issue comments, labels PM: - Labels issues during triage, comments triage status - Scans open PRs for external review requests - Comments on issues linking merged PRs Engineer: - Replies to review comments on PR after fixing - Reviews external PRs with APPROVE/REQUEST_CHANGES + line comments QA: - Leaves PR review (APPROVE or REQUEST_CHANGES with line comments) - All findings visible on GitHub, not just via agent_send SKILL.md: - Added PR diff, reviews, review comments, reply, merge API references * fix(hands/devteam): enforce English comments, line-level reviews, comment-before-close - All GitHub comments/reviews must be in English (added global rule) - PR reviews must use comments[] with path+line, not body-only - Comment on issue with resolution details BEFORE closing/merging - Improved comment templates with structured info * fix(hands/devteam): 8 logic fixes from end-to-end workflow review 1. Filter PRs from Issues API (pull_request key) 2. PM sends PR number to QA for review 3. Deduplicate PR scanning via devteam_reviewed_prs 4. QA reports test gaps instead of pushing code to shared branch 5. branch_strategy wired into Engineer (gitflow branches from develop) 6. approval_mode: ON = wait for human, OFF = auto-merge after QA 7. scan_interval mapped to schedule_create every_secs 8. git checkout -B instead of -b to handle existing branches * fix(hands/devteam): second-pass review — 6 more logic fixes 1. Engineer extracts PR number from create-PR API response 2. PM falls back to GitHub Contents API when shared repo not yet cloned 3. QA gets external PR review flow (was only on Engineer) 4. PM checks CI status + mergeable before merging 5. PM handles merge conflict (409) by sending back to Engineer to rebase 6. i18n approval_mode description synced with actual semantics * fix(hands/devteam): third-pass — runtime scenarios 1. Deduplicate cron schedule on daemon restart (check schedule_list first) 2. Max 3 review rounds before escalating to user (prevent infinite loop) 3. Clean working directory before switching tasks (git checkout -- . && git clean) 4. Add user direct commands (work on #42, status, review PR #50) 5. Pass tech_stack to Engineer in task delegation 6. Fix duplicate step numbering in Review Cycle * fix(hands/devteam): fourth-pass — state consistency and edge cases 1. QA force-syncs to remote branch (git checkout -B origin/branch) for force-push safety 2. Board sync step: reconcile with GitHub each scan cycle (catch external closes/merges) 3. Prune devteam_reviewed_prs of closed PRs, cap done list at 30 4. PM checks CI before sending to QA (don't waste QA on red builds) 5. Stop/cancel command: remove from board, comment on issue 6. Explicit rebase commands for Engineer (fetch + rebase + force-with-lease) * fix(hands/devteam): fifth-pass — crash prevention 1. Guard empty repo_url: stop and tell user to configure it 2. Add python3 to requires (all JSON parsing depends on it) 3. Engineer git config user.name/email on first clone (prevents commit rejection) 4. Explicit build/lint/test commands per tech stack (Rust/TS/Python/Go/Java/Swift) 5. event_publish on task completion so user gets notified 6. Global rule: check API HTTP status before parsing JSON * feat(hands/devteam): add gh CLI / MCP / curl API three-layer fallback - Add GitHub MCP integration (mcp_servers = ["github"]) - Add gh CLI as optional requirement (preferred over curl) - All 3 agents: gh > MCP > curl priority for GitHub operations - Add issue_tracker setting (github/linear/jira) - Add agent_list to shared tools - SKILL.md: add full gh CLI reference section - i18n: add issue_tracker translation * feat(hands/devteam): full MCP/integration/notification layer MCP allowlist: github, linear, jira, sentry, slack, discord - Sentry: Engineer reads crash reports/stack traces when fixing bugs - Slack/Discord: PM posts status updates (triaged, completed, QA results) - Linear/Jira: alternative issue trackers New settings: notify_channel (none/slack/discord), issue_tracker (github/linear/jira) New optional requires: npx (MCP runtime), SENTRY_AUTH_TOKEN PM prompt: notification section, channel-aware status posting Engineer prompt: Sentry context lookup for bug fixes i18n: added translations for new settings * feat(hands/devteam): workflows, onboarding, knowledge, standup, rollback Workflows (8 integrated): - PM: bug-triage, product-spec, weekly-report, incident-postmortem - Engineer: code-review, test-generation, refactor-plan, api-design - QA: code-review, test-generation New capabilities: - Repo onboarding: first activation analyzes repo structure/stack/CI - Knowledge accumulation: store lessons per issue, detect module hotspots - Daily standup: cron schedule, board summary via notify_channel - Rollback: gh pr revert + postmortem workflow + re-open issue Also: - Added workflow_run to tools, skills = [] (all allowed) - Rewrote README with full architecture, lifecycle, workflow table - PM prompt now has 15 sections covering full lifecycle * feat(hands/devteam): per-agent capabilities, resources, profiles, fallbacks Each agent now has full AgentManifest config (not just system_prompt): PM: - profile: automation - capabilities: web, memory, schedule, knowledge, event, workflow, agent_send - shell: gh, curl, cat, python3 - resources: 200k tokens/hr Engineer: - profile: coding - capabilities: file r/w, shell, web, memory, knowledge, workflow - shell: cargo, npm, python, go, swift, mvn, git, gh, docker, make - resources: 300k tokens/hr, 10 concurrent tools - network: * (needs to push to GitHub) QA: - profile: coding (read-heavy, no file_write) - capabilities: file read, shell (test/lint commands only), web, workflow - shell: cargo test/clippy/audit, npm test, pytest, go test, gh - resources: 150k tokens/hr All agents have fallback_models configured. * feat(hands/devteam): rewrite with proper resource composition First hand to use the new composition features: Agents: - PM: base=planner, capabilities restricted to gh/git shell only - Engineer: base=coder, full shell access, network=* - QA: base=code-reviewer, tool_blocklist=[file_write], test/lint shells only Composition: - base: inherit from agents/planner, agents/coder, agents/code-reviewer - mcp_servers: github (agents interact via MCP, not curl in prompts) - workflows: bug-triage, code-review, test-generation via workflow_run tool - plugins: todo-tracker, auto-summarizer, episodic-memory - per-agent skills: SKILL-pm.md, SKILL-engineer.md, SKILL-qa.md - per-agent capabilities: QA can't write files, PM can't run builds Prompts are clean and focused (role + methodology + principles), not stuffed with curl commands. GitHub interaction goes through MCP tools or gh CLI. * fix(devteam): complete planner methodology in PM prompt Added SCOPE/SEQUENCE/RISK/MILESTONE keywords from the planner base template's methodology into the PM's triage workflow. * docs: update hands README, fix repo_url reference in prompts - hands/README.md: document full composition model (base, MCP, workflows, plugins, per-agent skills, per-agent capabilities) - Updated hand count to 15 (added devteam) - Engineer prompt: clarify repo_url comes from User Configuration, not a template variable - PM prompt: same clarification * fix(devteam): override name/description from base templates Without explicit name, agents inherit base names (planner/coder/code-reviewer) instead of hand-specific names (pm/engineer/qa). This affects display and the prefixed name used in agent registry (devteam:pm vs devteam:planner). * style: format HAND.toml with taplo |
||
|
Evan Hu
|
33d279889c |
feat(hands): complete i18n fixes, SKILL.md enhancements, and README overhaul
- Fix French accent characters (é/è/ê/ç/â/ô) across all 14 HAND.toml files - Fix German special characters (ä/ö/ü/ß) across all 14 HAND.toml files - Add category translations to all 6 i18n language blocks in all 14 hands - Enhance SKILL.md content for 9 hands with practical examples and workflows - Trim bloated SKILL.md files (apitester 1400→892, devops 1301→870) - Rewrite root README.md with accurate stats, complete hand/integration tables - Update hands/README.md with full 14-hand listing and i18n documentation |
||
|
Evan Hu
|
206169c1d7 |
docs: add README for every content directory
Each directory (agents, hands, integrations, plugins, providers, scripts, skills) now has a README documenting its TOML format, current contents, and contribution steps. |