@krtclcdy/tersio 1.0.4 → 2.0.1
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.
- package/README.md +25 -10
- package/extensions/ai-addons-updater/index.js +166 -238
- package/extensions/caveman-session/index.js +53 -67
- package/extensions/combo-toggle/index.js +63 -76
- package/extensions/lib/utils.js +77 -24
- package/extensions/rtk-session/index.js +53 -63
- package/extensions/shared/mode-reinforcement.js +11 -14
- package/extensions/shared/plugin-settings.js +18 -16
- package/extensions/shared/session-state.js +77 -45
- package/package.json +6 -15
- package/tersio.js +678 -752
- package/extensions/amanai-reward/index.js +0 -24
- package/extensions/amanai-reward/pi.js +0 -32
package/README.md
CHANGED
|
@@ -1,12 +1,13 @@
|
|
|
1
1
|
# Tersio ✂️
|
|
2
2
|
|
|
3
|
-
[](https://www.npmjs.com/package/@krtclcdy/tersio)
|
|
4
|
-
[](https://nodejs.org)
|
|
3
|
+
[](https://www.npmjs.com/package/@krtclcdy/tersio)
|
|
4
|
+
[](https://github.com/KurutoDenzeru/tersio/releases)
|
|
5
|
+
[](./LICENSE)
|
|
6
|
+
[](https://nodejs.org)
|
|
7
7
|
[](https://github.com/KurutoDenzeru/tersio/actions)
|
|
8
|
+
[](https://github.com/KurutoDenzeru/tersio/issues)
|
|
8
9
|
|
|
9
|
-
Terse replies, compact shell output, and minimal code decisions for [Oh My Pi (OMP)](https://github.com/can1357/oh-my-pi)
|
|
10
|
+
Terse replies, compact shell output, and minimal code decisions for [Oh My Pi (OMP)](https://github.com/can1357/oh-my-pi).
|
|
10
11
|
|
|
11
12
|
## Install
|
|
12
13
|
|
|
@@ -55,9 +56,23 @@ Defaults apply to fresh sessions only — anything persisted with `/combo`, `/ca
|
|
|
55
56
|
| **Ponytail** | Minimal, YAGNI-oriented code decisions |
|
|
56
57
|
| **Combo** | Toggles all three at once. Presets: `off`, `medium`, `balanced`, `max` |
|
|
57
58
|
| **Updater** | `/ai-addons` checks and updates Ponytail, RTK, and Caveman in-session, with dry-run and backups |
|
|
58
|
-
| **Amanai reward detector** | Local notice when a final response contains a footer-shaped `AMANAI-GACHA-…` key. Never alters output, stores or sends the key, redeems it, or makes requests — redeem keys yourself in the Amanai billing dashboard |
|
|
59
59
|
|
|
60
|
-
User-level installs also register the package in `~/.omp/plugins` (visible in OMP **Settings → Plugins**),
|
|
60
|
+
User-level installs also register the package in `~/.omp/plugins` (visible in OMP **Settings → Plugins**), directly through the plugin manifest. Active modes reassert after Ponytail's prompt block on every top-level turn, including after history compaction.
|
|
61
|
+
|
|
62
|
+
## Benchmarks
|
|
63
|
+
|
|
64
|
+
Measured savings against the same workload without the modes. Token counts are real `o200k_base` BPE tokens (js-tiktoken), not chars÷4 estimates. Sample protocol, per-sample data, and caveats in [BENCHMARK.md](./BENCHMARK.md).
|
|
65
|
+
|
|
66
|
+
| Measured surface (n) | Baseline → Tersio | Δ |
|
|
67
|
+
|---|---|---|
|
|
68
|
+
| Terse reply, caveman lite — BPE tok (3 samples, p50) | 146 → 73 tok | **−52.6%** |
|
|
69
|
+
| Terse reply, caveman full — BPE tok (3 samples, p50) | 154 → 58 tok | **−62.3%** |
|
|
70
|
+
| Terse reply, caveman ultra — BPE tok (3 samples, p50) | 146 → 33 tok | **−78.6%** |
|
|
71
|
+
| Code diff, ponytail minimal-ladder — BPE tok (2 tasks) | 261–481 → 108–143 tok | **−58.6…−70.3%** |
|
|
72
|
+
| Shell output, rtk — `git status` — BPE tok | 32 → 21 tok | **−34.4%** |
|
|
73
|
+
| Shell output, rtk — repo `grep` — BPE tok | 537 → 469 tok | **−12.7%** |
|
|
74
|
+
|
|
75
|
+
Break-even math, balanced preset (caveman full + rtk + ponytail): `⌈overhead ÷ saving-per-turn⌉ = ⌈480 ÷ 306⌉ = 2 turns`, where a mixed turn (reply + `git status`-class command + code task) costs 944 tok baseline vs 638 tok with Tersio — a 0.68× ratio. Every turn after turn 2 nets ≈ −32%. RTK keeps diffs and failing-test output exact by design (−0.4% and −1.5% there), concentrating savings where noise lives.
|
|
61
76
|
|
|
62
77
|
## CLI
|
|
63
78
|
|
|
@@ -67,10 +82,10 @@ User-level installs also register the package in `~/.omp/plugins` (visible in OM
|
|
|
67
82
|
| `tersio update` | Refresh from the latest release |
|
|
68
83
|
| `tersio reinstall` | Fresh install, preserving the Ponytail package |
|
|
69
84
|
| `tersio doctor` | Check OMP, extension, Ponytail, and RTK health |
|
|
70
|
-
| `tersio uninstall` | Remove extensions and
|
|
85
|
+
| `tersio uninstall` | Remove extensions, registration, and the Ponytail plugin (`--keep-ponytail` keeps Ponytail; `--remove-rtk` also removes the RTK binary) |
|
|
71
86
|
| `tersio version` | Print version |
|
|
72
87
|
|
|
73
|
-
Flags: `--dry-run`, `--yes`/`-y`, `--verbose`, `--scope`, `--combo-default`/`--caveman-default`/`--rtk-default`. Legacy `--doctor` / `--uninstall` forms still work.
|
|
88
|
+
Flags: `--dry-run`, `--yes`/`-y`, `--verbose`, `--scope`, `--combo-default`/`--caveman-default`/`--rtk-default`/`--ponytail-default`. Legacy `--doctor` / `--uninstall` forms still work.
|
|
74
89
|
|
|
75
90
|
## Commands reference
|
|
76
91
|
|
|
@@ -136,7 +151,7 @@ Covers noisy commands (`git status`, `git diff`, `read`, `grep`, test, `tsc`, li
|
|
|
136
151
|
|
|
137
152
|
| What | Path |
|
|
138
153
|
|---|---|
|
|
139
|
-
| Caveman / RTK / Updater / Combo
|
|
154
|
+
| Caveman / RTK / Updater / Combo extensions | `~/.omp/agent/extensions/{caveman-session,rtk-session,ai-addons-updater,combo-toggle}/` |
|
|
140
155
|
| Ponytail package | `~/.omp/plugins/node_modules/@dietrichgebert/ponytail/` |
|
|
141
156
|
| RTK binary | `~/.bun/bin/rtk` (`rtk.exe` on Windows) |
|
|
142
157
|
| Extension registrations | `~/.omp/agent/config.yml` |
|