This file captures project learnings that persist across agent sessions. It serves as institutional memory so agents don't repeat mistakes or rediscover established patterns.
How to update this file: When you learn something that future agents should know — a pattern that works well, a mistake to avoid, a key decision — add it to the appropriate section below. Keep entries concise (one or two lines). Include dates for decisions. Do not remove entries unless they are explicitly obsolete.
- Agent-model delegation: Router (Sonnet) reads task → selects Custom Agent from
.github/agents/→ spawns sub-agent with correct model tier viatasktoolmodelparameter. Works well. - httptest.NewServer for mocking GitHub tarball endpoint in installer tests — lets unit tests cover install/update without network calls.
- gh issue create --milestone does NOT work; must use
gh api repos/.../issueswith-F milestone=<number>. - Unauthenticated tarball fetch fails for private repos — always set
Authorization: Bearer $GH_TOKENheader in HTTP requests to GitHub API.
Record important architectural and process decisions with rationale. Link to ADRs when they exist.
- 2025-07-18: ADR-004 —
teamwork validatecommand. Exits 0/1/2, supports--json/--quiet, validates config+state+handoffs+memory in one pass. Seedocs/decisions/004-validate-command-design.md. - 2025-07-18: ADR-005 —
teamwork installandteamwork updatecommands. Tarball fetch (needs GH_TOKEN for private repos), manifest-based conflict detection (SHA-256), framework vs starter file classification. Seedocs/decisions/005-install-update-design.md. - 2026-03-03:
gh-teamworkCLI extension created at JoshLuedeman/gh-teamwork. Wrapsteamwork install/teamwork updatebehindgh teamwork init/gh teamwork update. Falls back to Docker if binary not found. - 2026-03-03: GitHub milestone numbering: #1=Phase 2 Orchestration App (closed), #2=Phase 1 install/update (closed), #3=Phase 2 gh extension (closed), #4=Phase 3 GitHub App, #5=Phase 4 MCP Integration, #6=Backlog, #7=Phase 2.5 CLI Enhancements.
- 2026-03-05: Restructured to Copilot-native file system. Roles moved from
agents/roles/to.github/agents/*.agent.md(Custom Agents — selectable from Copilot dropdown). Workflows moved fromagents/workflows/to.github/skills/*/SKILL.md(Skills — invocable via/skill-name). Added.github/instructions/for path-specific auto-loaded guidelines. DeletedCLAUDE.mdand.cursorrules. Added 3 new agents: lint-agent, api-agent, dba-agent (15 total). Agent files use<!-- CUSTOMIZE -->placeholders for project-specific details;/setup-teamworkskill fills them in.teamwork updatenow cleans up deprecated files automatically. - 2026-03-03: Phase 2 milestone closed. All 4 planning issues (#1-#4) resolved. Context sharing fully addressed by handoff system. Memory, metrics mostly addressed; follow-up tasks in Phase 2.5 (#66-#78). Multi-repo: hub-spoke design planned with 7 sub-tasks.
Things agents frequently get wrong. Check this section before starting work.
- Do not use
gh issue create --milestone— it silently ignores the milestone. Usegh apiinstead.
Persistent feedback from code reviews that applies broadly, not just to a single PR.
- (No entries yet — add broadly applicable review feedback here)