Conversation
- AGENTS.md with repo architecture, build commands, code navigation - CLAUDE.md symlink to AGENTS.md - /implement skill: plan, implement, verify, commit, draft PR workflow - .claude/skills symlink to .agents/skills
- Agent should write tests for new/changed logic when feasible - Skip and notify developer when effort would be disproportionate
Twinsen81
force-pushed
the
add-agents-md-skill
branch
2 times, most recently
from
46cf52a to
6cc4b6c
Compare
zwarm
left a comment
zwarm
left a comment
There was a problem hiding this comment.
@Twinsen81 - All looks good. It followed along with the skill. I have suggested a few changes, but none are required. Looks like CI is failing though.
|
|
||
| | Skill | Trigger phrases | | ||
| |-------------|------------------------------------------------------------------------------------------------------| | ||
| | `implement` | "implement", "fix this", "work on this", "build this", "add feature", any implementation request | |
There was a problem hiding this comment.
@Twinsen81:
nit would it make sense to have two skills? One for diagnose and one for implement OR does the "plan" cover this? I was thinking of the situation where you want to do analysis before having AI jump into implement.
| - **NEVER create a PR without explicit permission** | ||
| - When asked to "fix" or "update" something, that does NOT imply permission to commit/push | ||
| - Always wait for explicit "commit", "push", or "create PR" commands | ||
|
|
There was a problem hiding this comment.
@Twinsen81
What if we added something for branches/rebases "Do not create branches, rebase, or modify git history unless explicitly asked." ?
| ### 3c: Unit Tests | ||
|
|
||
| ```bash | ||
| ./gradlew aztec:testRelease |
There was a problem hiding this comment.
Missing a ":" in front of aztec
|
|
||
| **Test rules:** | ||
| - NEVER weaken or remove assertions to make tests pass | ||
| - NEVER modify production code just to pass a test without permission |
There was a problem hiding this comment.
@Twinsen81
Can we combine with something like this: "Don’t make unrelated/incorrect changes to production code solely to appease a test. If test reveals a real issue, fix it."
2 participants
Maintenance
Adds agent guidance files and a unified
/implementskill for coding agents working in this repository.Files added:
AGENTS.md— Repository architecture overview, module structure, code navigation, build/test commands, common crash patternsCLAUDE.md— Symlink toAGENTS.md.agents/skills/implement/SKILL.md— End-to-end implementation workflow (plan -> implement -> verify -> commit -> draft PR).claude/skills— Symlink to.agents/skillsTest
add-agents-md-skillbranch