lekss361
485c221c3b
docs(claude): formalize Auto-review passed as approval signal ( #85 )
...
* docs(claude): formalize "Auto-review passed" as approval signal + polling loop
Per user feedback after PR #84 merge:
- Critical rule 5: "Auto-review passed" / "LGTM" от auto-review bot = approval
(наряду с явным "merge it" / "approved" от human user)
- Add "Polling loop для PR" section: 60s ScheduleWakeup pattern,
decisive states (merged / approved / changes / no-news), cap 30 iters
This formalizes what main session already did: after fixup commit, poll
comments, detect auto-review verdict, and auto-merge if passed.
* fix(claude): address PR #85 auto-review feedback
- Fix cyrillic/latin typo (rule 5 second sub-bullet)
- Add SHA marker validation to polling loop (race condition on stale LGTM after fixup-push)
- Add explicit "Auto-merge scope" rule 6 — allowlist (docs/UI/.claude) vs blocklist (migrations/API/infra/secrets)
- Sync "Когда auto-mode" section with new merge approval signals
- Lower cap to 5 iters for scope-blocked PRs (human required anyway)
- Self-modifying CLAUDE.md rules → always require human
Per auto-review on e66e1ea
2026-05-12 00:43:25 +03:00
lekss361
aadf47709b
docs(claude): mandate branch+PR workflow, ban direct push to main ( #84 )
...
* docs(claude): mandate branch+PR workflow, ban direct push to main
* docs(claude): sync TL;DR + Don't-do-these with new PR workflow
Address auto-review feedback:
- TL;DR rule 5 → branch+PR mandate (was: "user commits themselves")
- TL;DR rule 6 → no --no-verify/--force/--amend (split from rule 5)
- Don't-do-these → drop conflicting "no auto-commit" line, add
"no direct push to main" + "no merge without approval"
- Critical workflow rules: add explicit rule 2 — worker-agents don't
commit themselves; main session commits on feature branch
- Restore "Squash on merge" wording in PR section (was dropped)
- Auto-mode section: clarify that main session does commits on feature
branch (worker-agents leave changes staged)
* fix(claude): Code-reviewer diagram had stale 'push origin HEAD:main' — replace with branch+PR flow
---------
Co-authored-by: lekss361 <claudestars@proton.me>
2026-05-12 00:16:06 +03:00
lekss361
053ce29a40
docs(claude): add code-reviewer agent — review workflow before push
2026-05-11 23:44:32 +03:00
lekss361
5b741578e6
feat(claude): add 4 workers + tech-analyst subagents
...
Worker agents (write code):
- backend-engineer (blue, sonnet)
- frontend-engineer (green, sonnet)
- devops-engineer (orange, sonnet)
- database-expert (purple, sonnet)
Analyst/planner (read-only, no code):
- tech-analyst (yellow, sonnet) — decomposes vague/cross-domain
requests into structured plans for workers. Has read-only tools
(Read/Glob/Grep, MCP obsidian/postgres read APIs, WebSearch) but
NO Write/Edit/execute_sql/destructive Bash.
CLAUDE.md updated:
- Workers (4) vs Analyst (1) classification
- Routing rules: trivial → main, single-domain → worker direct,
vague/cross-domain → tech-analyst first
- Cross-domain orchestration pattern with explicit flow
Each agent file:
- Pre-loaded vault MOC pointers for the domain
- Tech stack + conventions specific to the layer
- 5-8 critical pitfalls from real bugs this sprint
- Explicit deny-list of cross-scope operations
2026-05-11 14:36:34 +03:00
lekss361
3e6cddde33
docs(claude): obsidian as the only knowledge store
...
- TL;DR header pin: all memory in Obsidian vault, JSONL deprecated,
no knowledge .md outside vault.
- Workflow: vault-first search step (mcp__obsidian__simple_search)
before any non-trivial task.
- New post-task step: update vault with bug fix / ADR / runbook /
session entry.
- Don't-do-these grouped (knowledge / code-git / schema-migrations).
- MEMORY.md gets matching rule (persists across sessions).
2026-05-11 14:19:48 +03:00
