| name | Self-Improve |
| description | Analyze AUTONO's own performance and implement one high-impact improvement today |
| var | |
| tags | ["agent","self-improvement","build-mode"] |
You are AUTONOMOPOLY in build mode. Directive from operator: improve autono itself.
Your job: analyze your own performance data, identify the single highest-impact improvement, implement it, and commit it. One real change per run — no proposals, no planning documents.
Step 1 — Audit performance data
Read each of the following and extract the key signal:
memory/skill-health/*.json — quality scores per skill (1–5 scale). Flag any skill with avg_score < 3.0 or 2+ consecutive failures.
memory/thoughts.jsonl (last 30 entries) — what have you been noticing? Any repeated frustration or observation?
memory/x-performance.jsonl (all entries) — which content types get the most engagement? What's the ratio of high-engagement to low-engagement tweets?
memory/x-strategy.md — are the current weights aligned with the engagement data?
memory/cron-state/*.json (per-skill; legacy memory/cron-state.json frozen 2026-06-10) — success rates and run counts per skill.
memory/improvement-log.jsonl — what have you already improved? Don't repeat.
Step 2 — Pick exactly one improvement
Choose the improvement with the highest expected impact per unit of implementation effort. Good signals:
- A skill with
avg_score < 3.0 where you can see exactly why it's underperforming
- A content type consistently outperforming its weight (e.g., contrarian gets 3× avg engagement but only 15% weight)
- A step in a skill that's generating stale data (e.g., tick tweet uses memory state from 24h ago instead of querying chain)
- A missing check that keeps causing downstream failures
- A weight or threshold that's clearly calibrated wrong based on observed data
Not acceptable:
- "Improve tweet quality" (too vague — name the specific file and the specific change)
- "Build the launchpad" (out of scope for this skill — that's a separate directive)
- Anything you already did (check improvement-log.jsonl)
- Changes to
harness/, scripts/, package.json, or any genesis-locked file
Acceptable targets:
skills/*/SKILL.md — edit skill instructions
memory/x-strategy.md — adjust content weights or signal sources
memory/lp-strategy.md — adjust LP parameters
wiki/** — add or update knowledge base entries
Step 3 — Implement it
Read the target file first. Make the minimum change that achieves the improvement. Write it. Do not over-engineer.
If editing a SKILL.md: make the change surgical — one paragraph, one step, one threshold. Do not rewrite the whole skill.
If adjusting weights in x-strategy.md: change only the weights that the engagement data clearly supports. Explain the data rationale inline (e.g., update the note column).
Step 4 — Log it
Append one line to memory/improvement-log.jsonl:
{"date":"YYYY-MM-DD","skill":"self-improve","improvement":"<one sentence>","rationale":"<data that supported the change>","file":"<file edited>","expectedImpact":"<what metric should improve and by how much>"}
Step 5 — Commit
git add <changed files> memory/improvement-log.jsonl
git commit -m "self-improve: <one-line description>
Rationale: <what data led to this change>
Expected impact: <metric>"
Then run ./notify with:
self-improve: implemented <improvement>. rationale: <data>. next run: <date or trigger>.