Lopi context

Skill konjoai/lopi/.claude/skills/lopi-context

Full lopi project context — complete phase plan, current health metrics, standing rules for agents, crate details. Auto-load when working on lopi sprint planning, architecture decisions, or Phase 5+ work.From its SKILL.md

Install

npx -y skills add konjoai/lopi --skill lopi-context

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.9 KB, as published. Nobody here has run it

lopi — Full Project Context

Phase Plan

Phases 1-9 below are the original v0 plan and are all long shipped or superseded. The live numbering is the P-series (review-pipeline sprints). CHANGELOG.md and LEDGER.md are authoritative for what shipped and why; NEXT_SESSION_PROMPT.md (newest first) is authoritative for what is next. Do not plan from this table.

Phase	Status	What Shipped
1 — MVP Core	✅ v0.1.0	Cargo workspace, CLI verbs
2 — N Parallel Agents + Live Dashboard	✅ v0.4.0	AgentPool, EventBus, ratatui TUI, web dashboard, WebSocket
3 — Remote Control + Pattern Mining	✅ v0.3.0	GitHub webhook HMAC, pattern miner (the Telegram transport shipped here and was removed again in S10 Phase 4)
4 — Scheduled Tasks + Repo Profiles	✅ v0.5.0	cron scheduler, RepoProfile, `lopi watch --remote`
5 — Self-Improvement Engine	✅ v0.10.0 / v0.11.0	`lopi learn`, failure post-mortem, pattern learning
6 — GitHub Webhooks + CI Integration	◐ partial	`lopi-webhook` end-to-end; GitHub App mode and full-event HMAC still open
7 — Production Web UI	✅	SvelteKit dashboard under `web/`
8 — Native Mobile App	◐	macOS/iOS app under `macos/`; React Native never started
9 — Intelligence + Evolution	◐ ongoing	Sprints H-S, then the P-series

Sprints H through S (v0.10.0-v0.19.0) and the P-series (P0-P4, v0.40.0-v0.45.0) all postdate this table. See PLAN.md for the alphabetic sprint log (itself current only to v0.19.0) and docs/LOOP_ENGINEERING_ROADMAP.md for the freshest verified roadmap.

Current Health

Re-measure rather than trusting these; they drift. Verified 2026-08-19 at 0.46.0.

Crates: 19 under crates/, plus the root binary at src/ and a SvelteKit front end at web/
Tests: ~2196 #[test]/#[tokio::test] sites across crates/ + src/
Coverage: 68.34% (.konjo/coverage-floor.txt), against a soft 80% gate
Ratchets at zero margin: function-length 74, indexing floor 211
CLI: 31 subcommands (see src/cli.rs) — run, sail, watch, learn, spec, index, mcp-serve and more
Gate tiers: only G1 · static and two G2 · coverage steps are BLOCKING; everything else is ADVISORY (LEDGER.md's Gate-Tiering-1)

Known unreachable code — real, documented, not hidden

The direct-Anthropic-API tier (with_api()) has no production call site; the CLI fallbacks (verifier_cli.rs, postmortem_cli.rs) cover for it. --adaptive-retry is inert on the shipped binary.
lopi-remote is unreachable (README says so outright).
lopi replay reports a plan; live re-execution is not wired.

Standing Rules for Agents on lopi

Stay inside crates/ and src/ — never touch .github/, infra/, root Cargo.lock deliberately
cargo build must stay green — if it goes red, fix before doing anything else
No unwrap() outside tests — use anyhow::Result and ?
No silent failure — if a fallback path swallows an error, log via tracing::warn!
Async everywhere — Tokio is the only runtime; no blocking I/O on async paths (use spawn_blocking)

Key Architecture Decisions (Load-Bearing)

EventBus<TaskStatus> uses tokio::broadcast — subscriber lag drops old events; this is intentional
AgentPool uses Arc<Semaphore> for bounded concurrency — no global mutex on the task queue
lopi-memory uses sqlx with SQLite + WAL mode — concurrent readers are fine, one writer at a time
Pattern mining: keyword fingerprint extraction via find_similar_patterns(goal) — not embedding-based yet (Phase 9)
GitHub webhook HMAC: constant-time comparison via subtle::ConstantTimeEq — do not simplify to ==