self-improving-agent / token-tracker · read-only MVP
Token
去哪了
17
episodes ·
13
action_types · generated
2026-06-03T07:37:06Z
UTC
build.py · single-file · self-contained
Overview
·
at a glance
Total episodes
17
across 13 action_types
Agent runs
4
1 success · 3 partial · 0 failed
Total tokens
115.4k
sum
11% episodes have data
Tool events
145
from hooks.log
Episode activity over time · by hour
06:00
14:00
16:00
By outcome
8 / 4 / 0
success / partial / failed
Token 流向
·
by action_type
agent_run
4 ep
promotion_verification
2 ep
policy_promotion
1 ep
meta_learning
1 ep
experiment
1 ep
milestone
1 ep
proposal_implementation
1 ep
dry_run_validation
1 ep
validation
1 ep
experiment_complete
1 ep
promotion
1 ep
_unknown
1 ep
meta_analysis
1 ep
By subagent_type
_none
13 ep
general-purpose
4 ep
Agent 表现
·
by subagent_type
subagent_type
episodes
agent_runs
success
avg duration
avg tokens
avg tool_uses
first → last
_none
13
0
—
—
—
—
07:05 → 16:00
general-purpose
4
4
1/4
10.4s
57.7k
1.0
06:45 → 06:59
Activity stream
·
all
action_type
_unknown
(1)
agent_run
(4)
dry_run_validation
(1)
experiment
(1)
experiment_complete
(1)
meta_analysis
(1)
meta_learning
(1)
milestone
(1)
policy_promotion
(1)
promotion
(1)
promotion_verification
(2)
proposal_implementation
(1)
validation
(1)
subagent_type
_none
(13)
general-purpose
(4)
outcome
_unknown
(1)
in_progress
(1)
partial
(4)
partial_success
(1)
success
(8)
success (after 2 iterations to discover real field names)
(1)
success (with one self-corrected bug)
(1)
timestamp
↓
action_type
subagent_type
duration
tokens
outcome
summary
Observations
·
heuristic
agent_run 全部由
general-purpose
派出（4 次），没有其他 subagent_type 被实际 dispatch 过
agent_run 中
2/4
缺 total_tokens、
3/4
缺 duration_ms — hook 还在收集中
action_type 分布 Top 3: agent_run
4
, promotion_verification
2
, policy_promotion
1
工具使用 Top 3: Bash
90
, Write
17
, Edit
11
agent_run 样本量仅
4
次，< 10；选型决策建议等 ≥ 20 / ≥ 3 subagent_type 再做