awtoll 0.1.1__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- awtoll-0.1.1/LICENSE +202 -0
- awtoll-0.1.1/PKG-INFO +166 -0
- awtoll-0.1.1/README.md +148 -0
- awtoll-0.1.1/awtoll/__init__.py +9 -0
- awtoll-0.1.1/awtoll/_doctor.py +118 -0
- awtoll-0.1.1/awtoll/analyze.py +212 -0
- awtoll-0.1.1/awtoll/cli.py +514 -0
- awtoll-0.1.1/awtoll/history.py +224 -0
- awtoll-0.1.1/awtoll/ledger.py +196 -0
- awtoll-0.1.1/awtoll/selftest.py +499 -0
- awtoll-0.1.1/awtoll/shapes.py +183 -0
- awtoll-0.1.1/awtoll/toll.py +117 -0
- awtoll-0.1.1/awtoll/transcripts.py +281 -0
- awtoll-0.1.1/awtoll.egg-info/PKG-INFO +166 -0
- awtoll-0.1.1/awtoll.egg-info/SOURCES.txt +20 -0
- awtoll-0.1.1/awtoll.egg-info/dependency_links.txt +1 -0
- awtoll-0.1.1/awtoll.egg-info/entry_points.txt +2 -0
- awtoll-0.1.1/awtoll.egg-info/requires.txt +6 -0
- awtoll-0.1.1/awtoll.egg-info/top_level.txt +1 -0
- awtoll-0.1.1/pyproject.toml +60 -0
- awtoll-0.1.1/setup.cfg +4 -0
- awtoll-0.1.1/tests/test_awtoll.py +396 -0
awtoll-0.1.1/LICENSE
ADDED
|
@@ -0,0 +1,202 @@
|
|
|
1
|
+
|
|
2
|
+
Apache License
|
|
3
|
+
Version 2.0, January 2004
|
|
4
|
+
http://www.apache.org/licenses/
|
|
5
|
+
|
|
6
|
+
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
|
|
7
|
+
|
|
8
|
+
1. Definitions.
|
|
9
|
+
|
|
10
|
+
"License" shall mean the terms and conditions for use, reproduction,
|
|
11
|
+
and distribution as defined by Sections 1 through 9 of this document.
|
|
12
|
+
|
|
13
|
+
"Licensor" shall mean the copyright owner or entity authorized by
|
|
14
|
+
the copyright owner that is granting the License.
|
|
15
|
+
|
|
16
|
+
"Legal Entity" shall mean the union of the acting entity and all
|
|
17
|
+
other entities that control, are controlled by, or are under common
|
|
18
|
+
control with that entity. For the purposes of this definition,
|
|
19
|
+
"control" means (i) the power, direct or indirect, to cause the
|
|
20
|
+
direction or management of such entity, whether by contract or
|
|
21
|
+
otherwise, or (ii) ownership of fifty percent (50%) or more of the
|
|
22
|
+
outstanding shares, or (iii) beneficial ownership of such entity.
|
|
23
|
+
|
|
24
|
+
"You" (or "Your") shall mean an individual or Legal Entity
|
|
25
|
+
exercising permissions granted by this License.
|
|
26
|
+
|
|
27
|
+
"Source" form shall mean the preferred form for making modifications,
|
|
28
|
+
including but not limited to software source code, documentation
|
|
29
|
+
source, and configuration files.
|
|
30
|
+
|
|
31
|
+
"Object" form shall mean any form resulting from mechanical
|
|
32
|
+
transformation or translation of a Source form, including but
|
|
33
|
+
not limited to compiled object code, generated documentation,
|
|
34
|
+
and conversions to other media types.
|
|
35
|
+
|
|
36
|
+
"Work" shall mean the work of authorship, whether in Source or
|
|
37
|
+
Object form, made available under the License, as indicated by a
|
|
38
|
+
copyright notice that is included in or attached to the work
|
|
39
|
+
(an example is provided in the Appendix below).
|
|
40
|
+
|
|
41
|
+
"Derivative Works" shall mean any work, whether in Source or Object
|
|
42
|
+
form, that is based on (or derived from) the Work and for which the
|
|
43
|
+
editorial revisions, annotations, elaborations, or other modifications
|
|
44
|
+
represent, as a whole, an original work of authorship. For the purposes
|
|
45
|
+
of this License, Derivative Works shall not include works that remain
|
|
46
|
+
separable from, or merely link (or bind by name) to the interfaces of,
|
|
47
|
+
the Work and Derivative Works thereof.
|
|
48
|
+
|
|
49
|
+
"Contribution" shall mean any work of authorship, including
|
|
50
|
+
the original version of the Work and any modifications or additions
|
|
51
|
+
to that Work or Derivative Works thereof, that is intentionally
|
|
52
|
+
submitted to Licensor for inclusion in the Work by the copyright owner
|
|
53
|
+
or by an individual or Legal Entity authorized to submit on behalf of
|
|
54
|
+
the copyright owner. For the purposes of this definition, "submitted"
|
|
55
|
+
means any form of electronic, verbal, or written communication sent
|
|
56
|
+
to the Licensor or its representatives, including but not limited to
|
|
57
|
+
communication on electronic mailing lists, source code control systems,
|
|
58
|
+
and issue tracking systems that are managed by, or on behalf of, the
|
|
59
|
+
Licensor for the purpose of discussing and improving the Work, but
|
|
60
|
+
excluding communication that is conspicuously marked or otherwise
|
|
61
|
+
designated in writing by the copyright owner as "Not a Contribution."
|
|
62
|
+
|
|
63
|
+
"Contributor" shall mean Licensor and any individual or Legal Entity
|
|
64
|
+
on behalf of whom a Contribution has been received by Licensor and
|
|
65
|
+
subsequently incorporated within the Work.
|
|
66
|
+
|
|
67
|
+
2. Grant of Copyright License. Subject to the terms and conditions of
|
|
68
|
+
this License, each Contributor hereby grants to You a perpetual,
|
|
69
|
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
|
70
|
+
copyright license to reproduce, prepare Derivative Works of,
|
|
71
|
+
publicly display, publicly perform, sublicense, and distribute the
|
|
72
|
+
Work and such Derivative Works in Source or Object form.
|
|
73
|
+
|
|
74
|
+
3. Grant of Patent License. Subject to the terms and conditions of
|
|
75
|
+
this License, each Contributor hereby grants to You a perpetual,
|
|
76
|
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
|
77
|
+
(except as stated in this section) patent license to make, have made,
|
|
78
|
+
use, offer to sell, sell, import, and otherwise transfer the Work,
|
|
79
|
+
where such license applies only to those patent claims licensable
|
|
80
|
+
by such Contributor that are necessarily infringed by their
|
|
81
|
+
Contribution(s) alone or by combination of their Contribution(s)
|
|
82
|
+
with the Work to which such Contribution(s) was submitted. If You
|
|
83
|
+
institute patent litigation against any entity (including a
|
|
84
|
+
cross-claim or counterclaim in a lawsuit) alleging that the Work
|
|
85
|
+
or a Contribution incorporated within the Work constitutes direct
|
|
86
|
+
or contributory patent infringement, then any patent licenses
|
|
87
|
+
granted to You under this License for that Work shall terminate
|
|
88
|
+
as of the date such litigation is filed.
|
|
89
|
+
|
|
90
|
+
4. Redistribution. You may reproduce and distribute copies of the
|
|
91
|
+
Work or Derivative Works thereof in any medium, with or without
|
|
92
|
+
modifications, and in Source or Object form, provided that You
|
|
93
|
+
meet the following conditions:
|
|
94
|
+
|
|
95
|
+
(a) You must give any other recipients of the Work or
|
|
96
|
+
Derivative Works a copy of this License; and
|
|
97
|
+
|
|
98
|
+
(b) You must cause any modified files to carry prominent notices
|
|
99
|
+
stating that You changed the files; and
|
|
100
|
+
|
|
101
|
+
(c) You must retain, in the Source form of any Derivative Works
|
|
102
|
+
that You distribute, all copyright, patent, trademark, and
|
|
103
|
+
attribution notices from the Source form of the Work,
|
|
104
|
+
excluding those notices that do not pertain to any part of
|
|
105
|
+
the Derivative Works; and
|
|
106
|
+
|
|
107
|
+
(d) If the Work includes a "NOTICE" text file as part of its
|
|
108
|
+
distribution, then any Derivative Works that You distribute must
|
|
109
|
+
include a readable copy of the attribution notices contained
|
|
110
|
+
within such NOTICE file, excluding those notices that do not
|
|
111
|
+
pertain to any part of the Derivative Works, in at least one
|
|
112
|
+
of the following places: within a NOTICE text file distributed
|
|
113
|
+
as part of the Derivative Works; within the Source form or
|
|
114
|
+
documentation, if provided along with the Derivative Works; or,
|
|
115
|
+
within a display generated by the Derivative Works, if and
|
|
116
|
+
wherever such third-party notices normally appear. The contents
|
|
117
|
+
of the NOTICE file are for informational purposes only and
|
|
118
|
+
do not modify the License. You may add Your own attribution
|
|
119
|
+
notices within Derivative Works that You distribute, alongside
|
|
120
|
+
or as an addendum to the NOTICE text from the Work, provided
|
|
121
|
+
that such additional attribution notices cannot be construed
|
|
122
|
+
as modifying the License.
|
|
123
|
+
|
|
124
|
+
You may add Your own copyright statement to Your modifications and
|
|
125
|
+
may provide additional or different license terms and conditions
|
|
126
|
+
for use, reproduction, or distribution of Your modifications, or
|
|
127
|
+
for any such Derivative Works as a whole, provided Your use,
|
|
128
|
+
reproduction, and distribution of the Work otherwise complies with
|
|
129
|
+
the conditions stated in this License.
|
|
130
|
+
|
|
131
|
+
5. Submission of Contributions. Unless You explicitly state otherwise,
|
|
132
|
+
any Contribution intentionally submitted for inclusion in the Work
|
|
133
|
+
by You to the Licensor shall be under the terms and conditions of
|
|
134
|
+
this License, without any additional terms or conditions.
|
|
135
|
+
Notwithstanding the above, nothing herein shall supersede or modify
|
|
136
|
+
the terms of any separate license agreement you may have executed
|
|
137
|
+
with Licensor regarding such Contributions.
|
|
138
|
+
|
|
139
|
+
6. Trademarks. This License does not grant permission to use the trade
|
|
140
|
+
names, trademarks, service marks, or product names of the Licensor,
|
|
141
|
+
except as required for reasonable and customary use in describing the
|
|
142
|
+
origin of the Work and reproducing the content of the NOTICE file.
|
|
143
|
+
|
|
144
|
+
7. Disclaimer of Warranty. Unless required by applicable law or
|
|
145
|
+
agreed to in writing, Licensor provides the Work (and each
|
|
146
|
+
Contributor provides its Contributions) on an "AS IS" BASIS,
|
|
147
|
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
|
|
148
|
+
implied, including, without limitation, any warranties or conditions
|
|
149
|
+
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
|
|
150
|
+
PARTICULAR PURPOSE. You are solely responsible for determining the
|
|
151
|
+
appropriateness of using or redistributing the Work and assume any
|
|
152
|
+
risks associated with Your exercise of permissions under this License.
|
|
153
|
+
|
|
154
|
+
8. Limitation of Liability. In no event and under no legal theory,
|
|
155
|
+
whether in tort (including negligence), contract, or otherwise,
|
|
156
|
+
unless required by applicable law (such as deliberate and grossly
|
|
157
|
+
negligent acts) or agreed to in writing, shall any Contributor be
|
|
158
|
+
liable to You for damages, including any direct, indirect, special,
|
|
159
|
+
incidental, or consequential damages of any character arising as a
|
|
160
|
+
result of this License or out of the use or inability to use the
|
|
161
|
+
Work (including but not limited to damages for loss of goodwill,
|
|
162
|
+
work stoppage, computer failure or malfunction, or any and all
|
|
163
|
+
other commercial damages or losses), even if such Contributor
|
|
164
|
+
has been advised of the possibility of such damages.
|
|
165
|
+
|
|
166
|
+
9. Accepting Warranty or Additional Liability. While redistributing
|
|
167
|
+
the Work or Derivative Works thereof, You may choose to offer,
|
|
168
|
+
and charge a fee for, acceptance of support, warranty, indemnity,
|
|
169
|
+
or other liability obligations and/or rights consistent with this
|
|
170
|
+
License. However, in accepting such obligations, You may act only
|
|
171
|
+
on Your own behalf and on Your sole responsibility, not on behalf
|
|
172
|
+
of any other Contributor, and only if You agree to indemnify,
|
|
173
|
+
defend, and hold each Contributor harmless for any liability
|
|
174
|
+
incurred by, or claims asserted against, such Contributor by reason
|
|
175
|
+
of your accepting any such warranty or additional liability.
|
|
176
|
+
|
|
177
|
+
END OF TERMS AND CONDITIONS
|
|
178
|
+
|
|
179
|
+
APPENDIX: How to apply the Apache License to your work.
|
|
180
|
+
|
|
181
|
+
To apply the Apache License to your work, attach the following
|
|
182
|
+
boilerplate notice, with the fields enclosed by brackets "[]"
|
|
183
|
+
replaced with your own identifying information. (Don't include
|
|
184
|
+
the brackets!) The text should be enclosed in the appropriate
|
|
185
|
+
comment syntax for the file format. We also recommend that a
|
|
186
|
+
file or class name and description of purpose be included on the
|
|
187
|
+
same "printed page" as the copyright notice for easier
|
|
188
|
+
identification within third-party archives.
|
|
189
|
+
|
|
190
|
+
Copyright [yyyy] [name of copyright owner]
|
|
191
|
+
|
|
192
|
+
Licensed under the Apache License, Version 2.0 (the "License");
|
|
193
|
+
you may not use this file except in compliance with the License.
|
|
194
|
+
You may obtain a copy of the License at
|
|
195
|
+
|
|
196
|
+
http://www.apache.org/licenses/LICENSE-2.0
|
|
197
|
+
|
|
198
|
+
Unless required by applicable law or agreed to in writing, software
|
|
199
|
+
distributed under the License is distributed on an "AS IS" BASIS,
|
|
200
|
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
201
|
+
See the License for the specific language governing permissions and
|
|
202
|
+
limitations under the License.
|
awtoll-0.1.1/PKG-INFO
ADDED
|
@@ -0,0 +1,166 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: awtoll
|
|
3
|
+
Version: 0.1.1
|
|
4
|
+
Summary: What every tool call costs you in context — measured from your own agent transcripts, not claimed in a README.
|
|
5
|
+
License: Apache-2.0
|
|
6
|
+
Project-URL: Homepage, https://github.com/Aitherium/awtoll
|
|
7
|
+
Project-URL: Documentation, https://github.com/Aitherium/awtoll#readme
|
|
8
|
+
Project-URL: Repository, https://github.com/Aitherium/awtoll.git
|
|
9
|
+
Project-URL: Issues, https://github.com/Aitherium/awtoll/issues
|
|
10
|
+
Requires-Python: >=3.10
|
|
11
|
+
Description-Content-Type: text/markdown
|
|
12
|
+
License-File: LICENSE
|
|
13
|
+
Provides-Extra: exact
|
|
14
|
+
Requires-Dist: tiktoken>=0.5.0; extra == "exact"
|
|
15
|
+
Provides-Extra: dev
|
|
16
|
+
Requires-Dist: pytest>=7.0; extra == "dev"
|
|
17
|
+
Dynamic: license-file
|
|
18
|
+
|
|
19
|
+
# awtoll
|
|
20
|
+
|
|
21
|
+
**Aither World Toll** — what every tool call costs you in context, measured from your own
|
|
22
|
+
agent transcripts rather than claimed in a README.
|
|
23
|
+
|
|
24
|
+
Every agent stack asserts that its search / graph / memory tool is cheaper than grepping
|
|
25
|
+
and re-reading files. Almost nobody measures it. The claim ends up in a code comment,
|
|
26
|
+
measured once during development and gated by nothing — so a tool can regress into costing
|
|
27
|
+
more than the thing it replaced and every signal stays green.
|
|
28
|
+
|
|
29
|
+
`awtoll` reads the transcripts already on your disk and prints what you actually paid.
|
|
30
|
+
|
|
31
|
+
```bash
|
|
32
|
+
pip install awtoll # zero dependencies
|
|
33
|
+
pip install 'awtoll[exact]' # + tiktoken, for exact counts instead of estimates
|
|
34
|
+
|
|
35
|
+
awtoll scan # toll table: what each tool shape cost
|
|
36
|
+
awtoll repeats # toll you paid more than once
|
|
37
|
+
awtoll versus 'awgraph callers foo' 'grep -rn foo .'
|
|
38
|
+
awtoll check # gate: expensive shapes decided, waste ratcheting down
|
|
39
|
+
```
|
|
40
|
+
|
|
41
|
+
## The unit
|
|
42
|
+
|
|
43
|
+
The **toll** is the number of tokens a tool call's *result* adds to your context. It is
|
|
44
|
+
deterministic, needs no API key, and can be recomputed from a transcript months later.
|
|
45
|
+
|
|
46
|
+
```
|
|
47
|
+
shape calls ok total median p90
|
|
48
|
+
--------------------------------------------------------------------------------------
|
|
49
|
+
$ sed 650 645 271,868 308 827
|
|
50
|
+
$ grep 938 925 200,441 142 476
|
|
51
|
+
$ python -c (inline) 1927 1800 199,267 67 219
|
|
52
|
+
Read 63 63 46,586 378 1,265
|
|
53
|
+
```
|
|
54
|
+
|
|
55
|
+
## The rule everything here is built around
|
|
56
|
+
|
|
57
|
+
> **A tool that answers nothing is the cheapest tool there is.**
|
|
58
|
+
|
|
59
|
+
Rank tools by cost alone and the winner is whichever one is broken. So an `ok` toll is
|
|
60
|
+
never averaged together with an `error`, `empty` or `truncated` one. Those are counted,
|
|
61
|
+
printed, and held out — and a shape whose calls are mostly not-ok is **flagged as
|
|
62
|
+
suspicious** rather than celebrated as cheap.
|
|
63
|
+
|
|
64
|
+
There is a sixth outcome, `opaque`, and it exists because of a real false positive on this
|
|
65
|
+
tool's first run: a result made of structured blocks carrying no text is *not* an empty
|
|
66
|
+
answer, it is a cost a transcript cannot see. Rendering it to `""` reported 36 perfectly
|
|
67
|
+
healthy tool-loading calls as broken. Opaque calls are excluded from the judgement, not
|
|
68
|
+
counted as failures.
|
|
69
|
+
|
|
70
|
+
## Repeats — the strongest savings signal
|
|
71
|
+
|
|
72
|
+
The most defensible number here is not a model of what some other tool *would* have
|
|
73
|
+
returned. It is the toll you demonstrably paid twice:
|
|
74
|
+
|
|
75
|
+
```
|
|
76
|
+
wasted x each shape / target
|
|
77
|
+
8,332 2 8,332 Read :: .../resume-all.md
|
|
78
|
+
3,675 8 525 Read :: .../swebench_awdk.py
|
|
79
|
+
```
|
|
80
|
+
|
|
81
|
+
Eight reads of one file in one session is 3,675 tokens of pure re-purchase. Repeats are
|
|
82
|
+
scoped to a single session on purpose — re-reading a file next week is a new question, not
|
|
83
|
+
waste, and counting it would flood the report.
|
|
84
|
+
|
|
85
|
+
## Amplification: the toll is not what it costs
|
|
86
|
+
|
|
87
|
+
Every turn re-sends the whole context, so a token a tool admits *early* is paid for again
|
|
88
|
+
on every turn after it. Measured across 12 real sessions: 1.67M tokens of tool results sat
|
|
89
|
+
under 205M cumulative prompt tokens — a 123x multiple.
|
|
90
|
+
|
|
91
|
+
That figure is **not** a denominator and awtoll refuses to print it as one. An earlier
|
|
92
|
+
version divided the two and displayed "1%", which reads as *tool output is a rounding
|
|
93
|
+
error* when it means the opposite.
|
|
94
|
+
|
|
95
|
+
## Counts are labelled
|
|
96
|
+
|
|
97
|
+
With `tiktoken` installed, counts are exact. Without it, awtoll uses **3.459 chars/token**
|
|
98
|
+
— measured across 7,690 real tool results (median 3.463; the two agree, so the constant is
|
|
99
|
+
stable) — and labels every number ESTIMATED. Note that the folk constant of 4.0 would
|
|
100
|
+
understate every toll by ~13%: tool output is denser than prose.
|
|
101
|
+
|
|
102
|
+
## Exit codes
|
|
103
|
+
|
|
104
|
+
| code | meaning |
|
|
105
|
+
|---|---|
|
|
106
|
+
| 0 | measured, nothing violated |
|
|
107
|
+
| 1 | a rule was violated |
|
|
108
|
+
| 2 | **could not judge** — no transcripts, no tool calls, unreadable ledger |
|
|
109
|
+
|
|
110
|
+
2 is never 0. A meter that found nothing and a meter that measured a healthy system look
|
|
111
|
+
identical unless the tool refuses to call silence a pass.
|
|
112
|
+
|
|
113
|
+
## The ledger
|
|
114
|
+
|
|
115
|
+
`awtoll check` turns the measurement into a gate. A toll table nobody decides anything
|
|
116
|
+
about is a dashboard: read twice, then never again.
|
|
117
|
+
|
|
118
|
+
```bash
|
|
119
|
+
awtoll check --init # writes awtoll.json pinned to today's waste ratio
|
|
120
|
+
awtoll check # gate
|
|
121
|
+
```
|
|
122
|
+
|
|
123
|
+
```json
|
|
124
|
+
{
|
|
125
|
+
"decide_above_tokens": 50000,
|
|
126
|
+
"waste_ratio_pin": 0.012,
|
|
127
|
+
"decisions": {
|
|
128
|
+
"$ sed": { "status": "keep", "reason": "reading file slices is the job; the alternative is a whole-file Read" },
|
|
129
|
+
"Read": { "status": "replace", "with": "awgraph context — 8 reads of one file in one session" }
|
|
130
|
+
}
|
|
131
|
+
}
|
|
132
|
+
```
|
|
133
|
+
|
|
134
|
+
| rule | asserts |
|
|
135
|
+
|---|---|
|
|
136
|
+
| TL001 | every decision has a known status (`replace`, `keep`, `watch`) |
|
|
137
|
+
| TL002 | every status carries its second field — a reasonless decision is a hole dressed as a decision |
|
|
138
|
+
| TL003 | every shape above the threshold is decided |
|
|
139
|
+
| TL004 | no duplicate decision rows |
|
|
140
|
+
| TL005 | the repeat-waste ratio is pinned and ratchets **down** only |
|
|
141
|
+
|
|
142
|
+
## Reading transcripts other than Claude Code's
|
|
143
|
+
|
|
144
|
+
The default root is `~/.claude/projects`. Point somewhere else with `--root` (repeatable,
|
|
145
|
+
accepts a file or a directory) or `AWTOLL_TRANSCRIPTS`. The parser wants JSONL records with
|
|
146
|
+
`tool_use` / `tool_result` blocks paired by `tool_use_id`.
|
|
147
|
+
|
|
148
|
+
Results are paired **by id, never by adjacency**. Interleaved calls and sub-agent turns mean
|
|
149
|
+
the next record is frequently not the answer to the previous call, and positional pairing
|
|
150
|
+
produces a table that looks entirely reasonable while attributing every cost to the wrong
|
|
151
|
+
tool.
|
|
152
|
+
|
|
153
|
+
## Proving it still works
|
|
154
|
+
|
|
155
|
+
```bash
|
|
156
|
+
awtoll self-test
|
|
157
|
+
```
|
|
158
|
+
|
|
159
|
+
Every arm asserts a positive (the rule fires on the broken shape) and, where the rule could
|
|
160
|
+
over-fire, a negative (it stays quiet on the healthy one). A rule that always fires is as
|
|
161
|
+
useless as one that never does — the first floods and gets switched off, the second is
|
|
162
|
+
decoration.
|
|
163
|
+
|
|
164
|
+
## License
|
|
165
|
+
|
|
166
|
+
Apache-2.0
|
awtoll-0.1.1/README.md
ADDED
|
@@ -0,0 +1,148 @@
|
|
|
1
|
+
# awtoll
|
|
2
|
+
|
|
3
|
+
**Aither World Toll** — what every tool call costs you in context, measured from your own
|
|
4
|
+
agent transcripts rather than claimed in a README.
|
|
5
|
+
|
|
6
|
+
Every agent stack asserts that its search / graph / memory tool is cheaper than grepping
|
|
7
|
+
and re-reading files. Almost nobody measures it. The claim ends up in a code comment,
|
|
8
|
+
measured once during development and gated by nothing — so a tool can regress into costing
|
|
9
|
+
more than the thing it replaced and every signal stays green.
|
|
10
|
+
|
|
11
|
+
`awtoll` reads the transcripts already on your disk and prints what you actually paid.
|
|
12
|
+
|
|
13
|
+
```bash
|
|
14
|
+
pip install awtoll # zero dependencies
|
|
15
|
+
pip install 'awtoll[exact]' # + tiktoken, for exact counts instead of estimates
|
|
16
|
+
|
|
17
|
+
awtoll scan # toll table: what each tool shape cost
|
|
18
|
+
awtoll repeats # toll you paid more than once
|
|
19
|
+
awtoll versus 'awgraph callers foo' 'grep -rn foo .'
|
|
20
|
+
awtoll check # gate: expensive shapes decided, waste ratcheting down
|
|
21
|
+
```
|
|
22
|
+
|
|
23
|
+
## The unit
|
|
24
|
+
|
|
25
|
+
The **toll** is the number of tokens a tool call's *result* adds to your context. It is
|
|
26
|
+
deterministic, needs no API key, and can be recomputed from a transcript months later.
|
|
27
|
+
|
|
28
|
+
```
|
|
29
|
+
shape calls ok total median p90
|
|
30
|
+
--------------------------------------------------------------------------------------
|
|
31
|
+
$ sed 650 645 271,868 308 827
|
|
32
|
+
$ grep 938 925 200,441 142 476
|
|
33
|
+
$ python -c (inline) 1927 1800 199,267 67 219
|
|
34
|
+
Read 63 63 46,586 378 1,265
|
|
35
|
+
```
|
|
36
|
+
|
|
37
|
+
## The rule everything here is built around
|
|
38
|
+
|
|
39
|
+
> **A tool that answers nothing is the cheapest tool there is.**
|
|
40
|
+
|
|
41
|
+
Rank tools by cost alone and the winner is whichever one is broken. So an `ok` toll is
|
|
42
|
+
never averaged together with an `error`, `empty` or `truncated` one. Those are counted,
|
|
43
|
+
printed, and held out — and a shape whose calls are mostly not-ok is **flagged as
|
|
44
|
+
suspicious** rather than celebrated as cheap.
|
|
45
|
+
|
|
46
|
+
There is a sixth outcome, `opaque`, and it exists because of a real false positive on this
|
|
47
|
+
tool's first run: a result made of structured blocks carrying no text is *not* an empty
|
|
48
|
+
answer, it is a cost a transcript cannot see. Rendering it to `""` reported 36 perfectly
|
|
49
|
+
healthy tool-loading calls as broken. Opaque calls are excluded from the judgement, not
|
|
50
|
+
counted as failures.
|
|
51
|
+
|
|
52
|
+
## Repeats — the strongest savings signal
|
|
53
|
+
|
|
54
|
+
The most defensible number here is not a model of what some other tool *would* have
|
|
55
|
+
returned. It is the toll you demonstrably paid twice:
|
|
56
|
+
|
|
57
|
+
```
|
|
58
|
+
wasted x each shape / target
|
|
59
|
+
8,332 2 8,332 Read :: .../resume-all.md
|
|
60
|
+
3,675 8 525 Read :: .../swebench_awdk.py
|
|
61
|
+
```
|
|
62
|
+
|
|
63
|
+
Eight reads of one file in one session is 3,675 tokens of pure re-purchase. Repeats are
|
|
64
|
+
scoped to a single session on purpose — re-reading a file next week is a new question, not
|
|
65
|
+
waste, and counting it would flood the report.
|
|
66
|
+
|
|
67
|
+
## Amplification: the toll is not what it costs
|
|
68
|
+
|
|
69
|
+
Every turn re-sends the whole context, so a token a tool admits *early* is paid for again
|
|
70
|
+
on every turn after it. Measured across 12 real sessions: 1.67M tokens of tool results sat
|
|
71
|
+
under 205M cumulative prompt tokens — a 123x multiple.
|
|
72
|
+
|
|
73
|
+
That figure is **not** a denominator and awtoll refuses to print it as one. An earlier
|
|
74
|
+
version divided the two and displayed "1%", which reads as *tool output is a rounding
|
|
75
|
+
error* when it means the opposite.
|
|
76
|
+
|
|
77
|
+
## Counts are labelled
|
|
78
|
+
|
|
79
|
+
With `tiktoken` installed, counts are exact. Without it, awtoll uses **3.459 chars/token**
|
|
80
|
+
— measured across 7,690 real tool results (median 3.463; the two agree, so the constant is
|
|
81
|
+
stable) — and labels every number ESTIMATED. Note that the folk constant of 4.0 would
|
|
82
|
+
understate every toll by ~13%: tool output is denser than prose.
|
|
83
|
+
|
|
84
|
+
## Exit codes
|
|
85
|
+
|
|
86
|
+
| code | meaning |
|
|
87
|
+
|---|---|
|
|
88
|
+
| 0 | measured, nothing violated |
|
|
89
|
+
| 1 | a rule was violated |
|
|
90
|
+
| 2 | **could not judge** — no transcripts, no tool calls, unreadable ledger |
|
|
91
|
+
|
|
92
|
+
2 is never 0. A meter that found nothing and a meter that measured a healthy system look
|
|
93
|
+
identical unless the tool refuses to call silence a pass.
|
|
94
|
+
|
|
95
|
+
## The ledger
|
|
96
|
+
|
|
97
|
+
`awtoll check` turns the measurement into a gate. A toll table nobody decides anything
|
|
98
|
+
about is a dashboard: read twice, then never again.
|
|
99
|
+
|
|
100
|
+
```bash
|
|
101
|
+
awtoll check --init # writes awtoll.json pinned to today's waste ratio
|
|
102
|
+
awtoll check # gate
|
|
103
|
+
```
|
|
104
|
+
|
|
105
|
+
```json
|
|
106
|
+
{
|
|
107
|
+
"decide_above_tokens": 50000,
|
|
108
|
+
"waste_ratio_pin": 0.012,
|
|
109
|
+
"decisions": {
|
|
110
|
+
"$ sed": { "status": "keep", "reason": "reading file slices is the job; the alternative is a whole-file Read" },
|
|
111
|
+
"Read": { "status": "replace", "with": "awgraph context — 8 reads of one file in one session" }
|
|
112
|
+
}
|
|
113
|
+
}
|
|
114
|
+
```
|
|
115
|
+
|
|
116
|
+
| rule | asserts |
|
|
117
|
+
|---|---|
|
|
118
|
+
| TL001 | every decision has a known status (`replace`, `keep`, `watch`) |
|
|
119
|
+
| TL002 | every status carries its second field — a reasonless decision is a hole dressed as a decision |
|
|
120
|
+
| TL003 | every shape above the threshold is decided |
|
|
121
|
+
| TL004 | no duplicate decision rows |
|
|
122
|
+
| TL005 | the repeat-waste ratio is pinned and ratchets **down** only |
|
|
123
|
+
|
|
124
|
+
## Reading transcripts other than Claude Code's
|
|
125
|
+
|
|
126
|
+
The default root is `~/.claude/projects`. Point somewhere else with `--root` (repeatable,
|
|
127
|
+
accepts a file or a directory) or `AWTOLL_TRANSCRIPTS`. The parser wants JSONL records with
|
|
128
|
+
`tool_use` / `tool_result` blocks paired by `tool_use_id`.
|
|
129
|
+
|
|
130
|
+
Results are paired **by id, never by adjacency**. Interleaved calls and sub-agent turns mean
|
|
131
|
+
the next record is frequently not the answer to the previous call, and positional pairing
|
|
132
|
+
produces a table that looks entirely reasonable while attributing every cost to the wrong
|
|
133
|
+
tool.
|
|
134
|
+
|
|
135
|
+
## Proving it still works
|
|
136
|
+
|
|
137
|
+
```bash
|
|
138
|
+
awtoll self-test
|
|
139
|
+
```
|
|
140
|
+
|
|
141
|
+
Every arm asserts a positive (the rule fires on the broken shape) and, where the rule could
|
|
142
|
+
over-fire, a negative (it stays quiet on the healthy one). A rule that always fires is as
|
|
143
|
+
useless as one that never does — the first floods and gets switched off, the second is
|
|
144
|
+
decoration.
|
|
145
|
+
|
|
146
|
+
## License
|
|
147
|
+
|
|
148
|
+
Apache-2.0
|
|
@@ -0,0 +1,118 @@
|
|
|
1
|
+
"""Stack-aware `doctor` for awtoll.
|
|
2
|
+
|
|
3
|
+
GENERATED BY gen_aw_doctor.py -- DO NOT EDIT.
|
|
4
|
+
Regenerate it with the generator named above; a hand-edit here is reverted by
|
|
5
|
+
the next run and fails the parity gate.
|
|
6
|
+
|
|
7
|
+
Why a doctor exists at all: the aw* bricks are designed to COMPOSE, so the
|
|
8
|
+
interesting failures live BETWEEN them. "awtoll is installed" is not the useful
|
|
9
|
+
fact -- "awtoll is installed and the thing it pairs with is not" is. This reports
|
|
10
|
+
the whole stack, not just itself.
|
|
11
|
+
|
|
12
|
+
stdlib only, on purpose: a diagnostic that cannot run because a dependency is
|
|
13
|
+
missing is worthless precisely when you need it.
|
|
14
|
+
"""
|
|
15
|
+
from __future__ import annotations
|
|
16
|
+
|
|
17
|
+
import importlib.util
|
|
18
|
+
import shutil
|
|
19
|
+
import sys
|
|
20
|
+
|
|
21
|
+
#: Frozen from AitherOS/config/ecosystem.yaml at generation time. A shipped
|
|
22
|
+
#: package cannot read the registry, and a doctor that guessed at the family
|
|
23
|
+
#: would go stale in silence. Regenerate to update.
|
|
24
|
+
SELF = 'awtoll'
|
|
25
|
+
FAMILY = ['awarena', 'awask', 'awbac', 'awbrowse', 'awdit', 'awevolve', 'awfind', 'awgit',
|
|
26
|
+
'awgraph', 'awiam', 'awkno', 'awm', 'awmail', 'awnboard', 'awnest', 'awnet',
|
|
27
|
+
'awprism', 'awreason', 'awrecover', 'awrecurse', 'awrelay', 'awrepl', 'awresearch',
|
|
28
|
+
'awrise', 'awrun', 'awscreen', 'awseal', 'awshare', 'awtunnel', 'awvision',
|
|
29
|
+
'awvoice', 'awwall']
|
|
30
|
+
PAIRS_WITH = ['awgit', 'awgraph']
|
|
31
|
+
|
|
32
|
+
|
|
33
|
+
def _installed(mod: str) -> "str | None":
|
|
34
|
+
"""Version if importable, else None. Never raises -- a broken sibling must
|
|
35
|
+
not take the diagnostic down with it."""
|
|
36
|
+
try:
|
|
37
|
+
if importlib.util.find_spec(mod) is None:
|
|
38
|
+
return None
|
|
39
|
+
except (ImportError, ValueError):
|
|
40
|
+
return None
|
|
41
|
+
try:
|
|
42
|
+
from importlib.metadata import PackageNotFoundError, version
|
|
43
|
+
try:
|
|
44
|
+
return version(mod)
|
|
45
|
+
except PackageNotFoundError:
|
|
46
|
+
return "installed"
|
|
47
|
+
except Exception:
|
|
48
|
+
return "installed"
|
|
49
|
+
|
|
50
|
+
|
|
51
|
+
def report(out=None) -> int:
|
|
52
|
+
"""Print the stack picture. 0 = this brick and its pairs are present."""
|
|
53
|
+
out = out or sys.stdout
|
|
54
|
+
print(f"{SELF} doctor", file=out)
|
|
55
|
+
|
|
56
|
+
mine = _installed(SELF)
|
|
57
|
+
print(f" self {SELF} {mine or 'NOT IMPORTABLE'}", file=out)
|
|
58
|
+
shim = shutil.which(SELF)
|
|
59
|
+
print(f" command {shim or 'not on PATH'}", file=out)
|
|
60
|
+
|
|
61
|
+
# The stack. Siblings this brick pairs with are called out separately,
|
|
62
|
+
# because a missing pair is a REASON, while a missing unrelated brick is
|
|
63
|
+
# just a fact about your machine.
|
|
64
|
+
missing_pairs, present = [], []
|
|
65
|
+
for name in FAMILY:
|
|
66
|
+
v = _installed(name)
|
|
67
|
+
if v:
|
|
68
|
+
present.append(name)
|
|
69
|
+
elif name in PAIRS_WITH:
|
|
70
|
+
missing_pairs.append(name)
|
|
71
|
+
print(f" stack {len(present)}/{len(FAMILY)} aw* packages installed",
|
|
72
|
+
file=out)
|
|
73
|
+
if present:
|
|
74
|
+
print(f" {' '.join(sorted(present))}", file=out)
|
|
75
|
+
|
|
76
|
+
local = _local_checks()
|
|
77
|
+
for line in local:
|
|
78
|
+
print(f" {line}", file=out)
|
|
79
|
+
|
|
80
|
+
if mine is None:
|
|
81
|
+
print(f"\nverdict: {SELF} itself is not importable. Reinstall it before "
|
|
82
|
+
f"anything else here means much.", file=out)
|
|
83
|
+
return 1
|
|
84
|
+
if missing_pairs:
|
|
85
|
+
print(f"\nverdict: {SELF} works, but pairs with "
|
|
86
|
+
f"{', '.join(sorted(missing_pairs))} which "
|
|
87
|
+
f"{'is' if len(missing_pairs) == 1 else 'are'} not installed. "
|
|
88
|
+
f"That is a capability you are missing, not an error.", file=out)
|
|
89
|
+
return 0
|
|
90
|
+
print(f"\nverdict: {SELF} and everything it pairs with are present.", file=out)
|
|
91
|
+
return 0
|
|
92
|
+
|
|
93
|
+
|
|
94
|
+
def _local_checks() -> "list[str]":
|
|
95
|
+
"""Per-brick checks, if this package defines them.
|
|
96
|
+
|
|
97
|
+
Kept as a HOOK rather than generated guesses: the generator knows the family
|
|
98
|
+
from the registry, but it does not know what awtoll needs at runtime, and a
|
|
99
|
+
doctor that invented config requirements would be confidently wrong. A
|
|
100
|
+
package supplies `_doctor_local()` returning display lines; absent, the
|
|
101
|
+
stack picture above still stands on its own.
|
|
102
|
+
"""
|
|
103
|
+
try:
|
|
104
|
+
mod = importlib.import_module(f"{SELF}.doctor_local")
|
|
105
|
+
except Exception:
|
|
106
|
+
return []
|
|
107
|
+
try:
|
|
108
|
+
return list(mod._doctor_local())
|
|
109
|
+
except Exception as exc: # noqa: BLE001
|
|
110
|
+
return [f"local checks raised {type(exc).__name__}: {exc}"]
|
|
111
|
+
|
|
112
|
+
|
|
113
|
+
def main(argv: "list[str] | None" = None) -> int:
|
|
114
|
+
return report()
|
|
115
|
+
|
|
116
|
+
|
|
117
|
+
if __name__ == "__main__":
|
|
118
|
+
raise SystemExit(main())
|