← ClaudeAtlas

etemaro-core-agent-logs-reviewlisted

Review the Etemaro core agent's run directory (packages/core — state.json, lessons.json, decision-log.json, logs/agent-*.log, logs/actions-*.jsonl, logs/structured-*.jsonl) and deliver a self-contained HTML performance, strategy and stability report. Use whenever the user asks to review, analyse, audit or report on an Etemaro agent run or an instance under data/instances/ or data/remote-server/data/instances/ — "review the agent logs", "how did agent X perform", "why did the agent lose money", "analyse the run", "check for errors in the agent logs", "make a report from the data dir", "pull the logs and analyse them", "review remote-server". Also triggers on a handed-over agent data directory, on questions about settlement status, win rate, exit reasons, liveness or smart-wallet gate integrity, and when only the words "agent", "instance" or "run" plus a directory are given. Does NOT cover the copy-trader CLI bot — that is etemaro-copy-trader-logs-review.
romankurnovskii/etemaro · ★ 43 · AI & Automation · score 67
Install: claude install-skill romankurnovskii/etemaro
# Etemaro Core-Agent Run Review → HTML Report Produce an exhaustive review of an Etemaro **core agent** run and deliver a self-contained HTML report (inline CSS, email-safe) that mirrors `references/report-template.html`. Scope is deliberately narrow: this skill reads **only** the core-agent family written by `packages/core`. The copy-trader CLI bot (`apps/bot-copy-trader`, streams `copytrader_runtime` / `copytrader_wallet`) is a different family with different dialects — use `etemaro-copy-trader-logs-review` for it. Reading one family's log with the other's probe fails **silently** and yields an empty report. ## The one rule that matters most **Errors are not in `agent-*.log`.** The text log's bracket is a *component tag, not a severity level*; it carries essentially no error signal. Errors live in `structured-*.jsonl` (`api_error` / `swap_error` / `tx_error`) and `actions-*.jsonl` (`success: false`). A forensics pass that greps the text log finds nothing and falsely reports a clean run. Read every stream through `scripts/_dialects.py`. Never hand-write a `python3 -c` one-liner over a log file and never raw-grep one. ```python from _dialects import coverage, iter_events, cron_cycles, action_calls coverage(instance_dir) # ALWAYS first for ev in iter_events(instance_dir, window=(frm, to)): ... # ev.ts, ev.dialect, ev.component, ev.agent_id, ev.msg, ev.fields ``` `parsed: 0` with `total_lines > 0` is a **broken ingestion, not a quiet run** — rep