transcripto 0.1.0__py3-none-any.whl

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.
@@ -0,0 +1,229 @@
1
+ Metadata-Version: 2.4
2
+ Name: transcripto
3
+ Version: 0.1.0
4
+ Summary: Search everything your coding agents ever did, grade your own prompts, and price your decisions. Local, stdlib-only, your data never leaves the machine.
5
+ Author: Oscar Morke
6
+ License: MIT
7
+ Project-URL: Homepage, https://github.com/Morkeeth/transcripto
8
+ Project-URL: Source, https://github.com/Morkeeth/transcripto
9
+ Keywords: claude-code,coding-agents,transcripts,local-first,analytics
10
+ Classifier: Environment :: Console
11
+ Classifier: Programming Language :: Python :: 3
12
+ Classifier: License :: OSI Approved :: MIT License
13
+ Classifier: Topic :: Utilities
14
+ Requires-Python: >=3.9
15
+ Description-Content-Type: text/markdown
16
+ License-File: LICENSE
17
+ Dynamic: license-file
18
+
19
+ # transcripto
20
+
21
+ **your coding agents keep a transcript of every session. it is the most valuable
22
+ dataset you own, and you cannot scroll back far enough to read it. transcripto
23
+ indexes it, keeps only the turns you actually typed, and grades them.**
24
+
25
+ one command, no account, no signup, no cloud. it reads files that are already on
26
+ your disk and never opens a socket.
27
+
28
+ ```
29
+ uvx transcripto coach
30
+ ```
31
+
32
+ ## what you get back
33
+
34
+ this is a real run on one machine, pasted unedited:
35
+
36
+ ```
37
+ YOUR PROMPT HABITS, GRADED (offline, your machine only)
38
+
39
+ harness: claude
40
+ corpus : 2713 transcript(s), 366,664 records
41
+ kept : 3605 prompts you actually typed (0.98% of records)
42
+ ranked : 1911 episodes, 890 survived (47%)
43
+ tiers : commit 356 | write/edit 534 | reverted 2 | nothing durable 1019
44
+
45
+ SURVIVAL IS A PROXY: survival = a durable Write/Edit or an un-reverted git commit in-episode. A PROXY, not proof the work was correct or shipped.
46
+
47
+ SURVIVES MOST do more of these:
48
+ 65% (164/253) detailed (>40 words)
49
+ 64% (66/103) states-a-check-or-done-condition
50
+ 59% (404/690) intent:CHANGE
51
+ 58% (21/36) no-object (pronoun/vague)
52
+ 58% (98/170) cites-a-file-or-path
53
+
54
+ SURVIVES LEAST these tend to loop:
55
+ 33% (4/12) intent:REVERT
56
+ 39% (403/1030) intent:none
57
+ 40% (4/10) intent:TEST
58
+ 41% (293/709) terse (<8 words)
59
+ 44% (75/169) intent:DESCRIBE
60
+
61
+ + your best landed prompt, with its witness:
62
+ "mTERMINAL 8 — Mountain of Helicon · ~/CODE/mountain-of-helicon Read ~/CODE/mountain-of-helicon. Two…"
63
+ COMMIT-WITNESSED: git commit · corrections: 0
64
+
65
+ - your worst looped prompt, with its witness:
66
+ "ok, and lets see they might solve it in the future so i can go back to my beloeved routine :) Befor…"
67
+ NO-DURABLE-RECORD: read-only Bash only, no file change · corrections: 15 · assistant turns: 221
68
+ ```
69
+
70
+ those are my numbers. yours will be different, which is the whole point. the last
71
+ two lines are the ones that sting: it hands you back your own best and worst
72
+ prompt, verbatim, with the receipt for why it scored each one.
73
+
74
+ on that machine, prompts that wrote down what done looks like survived **64% of
75
+ the time (66 of 103)**. prompts with no stated intent survived **39% (403 of
76
+ 1030)**. i had spent a year blaming the model.
77
+
78
+ ## the proxy caveat, which travels with every number
79
+
80
+ an episode "survived" if a Write or Edit landed, or a git commit ran and nothing
81
+ reverted it inside the same transcript.
82
+
83
+ that is a durable **keystroke**, not a durable **outcome**. a commit is not proof
84
+ the code was right. a revert in a later session is invisible to it. a prompt
85
+ whose payoff was a decision rather than an edit reads as dead. it is a coaching
86
+ signal, not a verdict. if it ever prints something that flatters you, distrust it.
87
+
88
+ the caveat is printed in the output itself, every run, on purpose.
89
+
90
+ ## why your own gate matters here
91
+
92
+ at fleet scale roughly 95% of the `type: user` records in a transcript are not
93
+ you. they are tool results, injected skill bodies, sub-agent prompts, and
94
+ messages from other terminals, all wearing your role. transcripto gates on
95
+ `promptSource` (typed/queued, no meta, no sidechain) so it grades what you typed.
96
+
97
+ you can watch the gate do work: in the run above, 3605 of 366,664 records
98
+ survived it. that is 0.98%.
99
+
100
+ the same gate is what makes `cost` produce a number a spend tracker cannot:
101
+
102
+ ```
103
+ cost per human decision last 30 days 2026-07-28 → 2026-08-27
104
+
105
+ API-equivalent spend $8,892.49
106
+ your decisions 2934 turns you actually typed (promptSource typed/queued)
107
+ ────────────────────────────────────────────────────
108
+ cost per human decision $3.03
109
+
110
+ 46.4k agent messages · 16 per decision · 11.5B tokens · 149 sessions
111
+ 57.7k raw `type: user` records in the same window. dividing by those instead
112
+ would read $0.15, 19.7x too cheap.
113
+ ```
114
+
115
+ it prints both, so the gate's effect is something you can check rather than
116
+ something i am asserting. these are API-equivalent dollars at list rates,
117
+ because a transcript has no cost field, only token counts. on a subscription you
118
+ did not pay this.
119
+
120
+ ## honest limits
121
+
122
+ read these before you quote a number at anyone.
123
+
124
+ - **survival is a proxy**, described above. durable keystroke, not durable outcome.
125
+ - **one operator's corpus.** every figure in this README comes from one machine.
126
+ it is an existence proof that the measurement runs, not a finding about how
127
+ people prompt. run it on yours and you get yours.
128
+ - **two harnesses today: Claude Code and Codex.** nothing else is supported.
129
+ cursor, aider, and the rest are not read.
130
+ - **the habit labels are heuristics.** "states-a-check-or-done-condition" is a
131
+ pattern match over your text, not comprehension. it will misfile some prompts.
132
+ - **correlation, not instruction.** detailed prompts surviving more often does not
133
+ prove that padding a prompt causes survival.
134
+
135
+ ## privacy
136
+
137
+ it runs locally and never touches the network. the entire import list is
138
+ `sys, os, json, glob, re, sqlite3, argparse, datetime`. there is no socket, no
139
+ urllib, no requests, no subprocess, no telemetry, no analytics, and no account.
140
+ you can check that in one grep of the single file it ships as.
141
+
142
+ your transcripts stay in `~/.claude` and `~/.codex`. the index it builds stays in
143
+ `~/.trace`.
144
+
145
+ ## the rest of it
146
+
147
+ ```
148
+ transcripto index build / refresh (incremental)
149
+ transcripto watch live, new sessions get picked up as your agents work
150
+ transcripto ask YOUR OWN messages about a topic, newest first + a rollup
151
+ transcripto search full-text across everything (you + agents + tool logs)
152
+ transcripto find every session that wrote / edited / read a file
153
+ transcripto sessions recent sessions + their opening ask
154
+ transcripto stats what you actually work on
155
+ transcripto cost what ONE of your decisions costs
156
+ transcripto coach which of YOUR prompt habits survive (a proxy)
157
+ ```
158
+
159
+ `ask` is the one that kills "wait, did i lose something?". it answers "what was i
160
+ thinking about X across ALL my sessions", in your own words only.
161
+
162
+ ```
163
+ $ transcripto find USER-JOURNEY.md
164
+ 2026-08-20 WROTE ~/CODE/mountain-of-helicon-main/USER-JOURNEY.md
165
+ ```
166
+
167
+ the file you lost, found across every session you ever ran, one line.
168
+
169
+ ## Codex
170
+
171
+ ```
172
+ transcripto coach --harness codex
173
+ ```
174
+
175
+ reads `~/.codex` (sessions + archived_sessions), normalises it into the same rows,
176
+ and applies the identical survival proxy. it also ingests `history.jsonl` purely
177
+ as a control on the gate: it reports how many of its input lines also show up as
178
+ typed rollout turns, so you can see the gate agreeing with a second source.
179
+
180
+ ## install
181
+
182
+ ```
183
+ uvx transcripto coach
184
+ ```
185
+
186
+ no install, nothing to set up. or put it on your PATH:
187
+
188
+ ```
189
+ pipx install transcripto
190
+ ```
191
+
192
+ or run the single file with no packaging at all:
193
+
194
+ ```
195
+ git clone https://github.com/Morkeeth/transcripto
196
+ cd transcripto
197
+ python3 transcripto.py coach
198
+ ```
199
+
200
+ no dependencies, stdlib only, one file. the packaging adds nothing at runtime, it
201
+ just gives the file a name on your PATH. `trace` is a second name for the same
202
+ command.
203
+
204
+ ## tests
205
+
206
+ ```
207
+ ./test_coach.sh 14 assertions
208
+ ./test_codex.sh 14 assertions
209
+ ./test_cost.sh 12 assertions
210
+ ```
211
+
212
+ offline, no keys, on fixtures that inherit the real transcript shape including
213
+ all four ways a non-human record disguises itself as `type: user`.
214
+
215
+ the load-bearing one in `test_coach.sh` is `REVERTED IS NOT SURVIVED`: a commit
216
+ that got `reset --hard` in the same session left no durable record. flip that one
217
+ line and the suite goes red, which is the point. a generous proxy is a broken one.
218
+
219
+ ## why though
220
+
221
+ your agent history is proof. every "yeah it's done" has a real trace sitting
222
+ behind it. transcripto is the index that makes it checkable.
223
+
224
+ it is the fuel layer. on top of it you check what your agents *claim* against what
225
+ the trace *shows*, which is [mountain of helicon](https://github.com/Morkeeth/mountain-of-helicon).
226
+ the pitch was never "search your history". it is *prove your agent did what it
227
+ said, from your own local traces.*
228
+
229
+ local, MIT, no telemetry. star it if it finds you something you'd lost ™
@@ -0,0 +1,7 @@
1
+ transcripto.py,sha256=gITFpzqFWGu6JhxX9Wamh6hIfCKbwijlGe3CoEDz44g,63747
2
+ transcripto-0.1.0.dist-info/licenses/LICENSE,sha256=sNIV-_XXsBt96tiCGvGjrYcs8-Pf2gt0rD2nMG8RBg0,1068
3
+ transcripto-0.1.0.dist-info/METADATA,sha256=66pS-LMY3s4TCTbSxQy5l1tYhAD84-XS1pZmoNEb7IQ,8954
4
+ transcripto-0.1.0.dist-info/WHEEL,sha256=YVMoNqKzERt-wjUZwJ33xBGAwnFl-4cqbYkTtWa4itE,91
5
+ transcripto-0.1.0.dist-info/entry_points.txt,sha256=n08_q2ciXNUidMox8xlmCq3slv1aS4NGxGSCIJOYKOI,49
6
+ transcripto-0.1.0.dist-info/top_level.txt,sha256=9WcZWehnabAXzqsCQdshLVAn9n6-8ttFf3_fzrbsAC4,12
7
+ transcripto-0.1.0.dist-info/RECORD,,
@@ -0,0 +1,5 @@
1
+ Wheel-Version: 1.0
2
+ Generator: setuptools (84.0.0)
3
+ Root-Is-Purelib: true
4
+ Tag: py3-none-any
5
+
@@ -0,0 +1,2 @@
1
+ [console_scripts]
2
+ transcripto = transcripto:main
@@ -0,0 +1,21 @@
1
+ MIT License
2
+
3
+ Copyright (c) 2026 Oscar Morke
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy
6
+ of this software and associated documentation files (the "Software"), to deal
7
+ in the Software without restriction, including without limitation the rights
8
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9
+ copies of the Software, and to permit persons to whom the Software is
10
+ furnished to do so, subject to the following conditions:
11
+
12
+ The above copyright notice and this permission notice shall be included in all
13
+ copies or substantial portions of the Software.
14
+
15
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
+ SOFTWARE.
@@ -0,0 +1 @@
1
+ transcripto