tomorrow: top 5 actions
prioritized for immediate impact. do these before anything else.
1. add file references to every thread opener
why: +25pp success rate. strongest single predictor in the dataset.
how: start threads with @path/to/file.ts instead of abstract descriptions.
- bad: “the auth system needs work”
- good: “@src/auth/login.ts needs error handling for expired tokens”
time: 0 minutes — just do it.
2. add confirmation gates to AGENTS.md
why: 47% of your steerings are “no…” and “wait…” — most correct premature agent action.
copy-paste this:
## confirmation gates
confirm with user before:
- running full test suites or benchmarks
- pushing code or making commits
- modifying files outside explicitly mentioned scope
- making architectural decisions
ask: "ready to run the tests?" rather than "running the tests now..."
time: 2 minutes.
3. schedule implementation work for 14:00-17:00
why: your evening sessions (19:00-22:00) have 20% resolution. afternoon peaks at 60%.
action plan:
- block 14:00-17:00 tomorrow for critical implementation
- evening = exploration, reading, handoff setup only
- if something must happen at night, make it research threads
time: 5 minutes calendar adjustment.
4. commit to 25+ turns before bailing
why: 54.6% of your threads die before turn 15. those resolve at 14%. threads at 26-50 turns resolve at 75%.
practical rule: before opening a thread, ask “will i stay 15+ turns?” if no, batch it with other tasks or don’t start.
minimum commitment: 25 turns for implementation work, 15 for exploration.
time: behavioral — no setup.
5. explicit approvals every 5 messages
why: your 0.55 approvals/thread vs 1.54 for power users. approval starvation causes agent drift.
action: after correct work, say “good” / “continue” / “ship it”. literally those words.
target: 1 approval per 5 agent messages. 2:1 approval:steering ratio.
time: behavioral — no setup.
verification
after implementing, track these metrics for one week:
| metric | current | target |
|---|---|---|
| threads with file refs in opener | ~30% | 60% |
| afternoon implementation sessions | ? | 3+ |
| threads reaching turn 15 | 45% | 55% |
| approvals per thread | 0.55 | 0.85 |
compiled from 4,656 thread analysis | 2026-01-09