loadout-ai 0.9.3 → 0.9.5

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/docs/REFERENCE.md CHANGED
@@ -42,10 +42,14 @@ pinned public sources**, installed into each agent you choose.
42
42
  | [Agent Skills Marketplace](https://github.com/wshobson/agents) | Architecture, review, error handling, JavaScript, Python | [![GitHub stars](https://img.shields.io/github/stars/wshobson/agents?style=flat&label=stars)](https://github.com/wshobson/agents) |
43
43
 
44
44
  ```bash
45
- loadout setup --mode stable
46
- loadout setup --mode stable --yes
45
+ loadout setup --mode stable --details
46
+ loadout setup --mode stable --yes --approve-risk
47
47
  ```
48
48
 
49
+ The current Stable selection reports instruction-like files in reviewed skill
50
+ sources. Read the detailed preview before acknowledging those findings with
51
+ `--approve-risk`.
52
+
49
53
  ### Power: a larger cross-project toolkit
50
54
 
51
55
  Power draws a skill-level allowlist from eight major collections.
@@ -107,11 +107,12 @@ explicit.
107
107
  For unattended use only after reviewing a preview, the equivalent is:
108
108
 
109
109
  ```bash
110
- loadout setup --mode stable --agents codex,claude-code --yes
110
+ loadout setup --mode stable --agents codex,claude-code --yes --approve-risk
111
111
  ```
112
112
 
113
- Do not add `--approve-risk` unless the displayed preview identifies a specific
114
- reviewed finding and you understand it.
113
+ The current Stable preview identifies instruction-like files in reviewed skill
114
+ sources, so unattended apply requires `--approve-risk`. Add it only after reading
115
+ and accepting the displayed findings.
115
116
 
116
117
  ## 5. Test a change and recover
117
118
 
@@ -0,0 +1,65 @@
1
+ # Fast Launch Demo Implementation Plan
2
+
3
+ > **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
4
+
5
+ **Goal:** Replace the five-minute demo script with a deterministic 60–75 second narrated launch cut.
6
+
7
+ **Architecture:** Prepare genuine CLI outputs and Kokoro narration before capture, replay those artifacts on a fixed scene timeline, then mux and verify one 1080p MP4. The completed discussion thread `a71874c7-5ac` supplies real Claude Code and Codex output without paid-turn waiting.
8
+
9
+ **Tech Stack:** Bash, Loadout CLI, Kokoro, macOS `screencapture`, FFmpeg, FFprobe.
10
+
11
+ ## Global Constraints
12
+
13
+ - Preserve Maximum, Optimize, Handoff, Coordination, and Discussion scenes.
14
+ - Omit rollback.
15
+ - Do not fabricate provider output.
16
+ - Cap the final video at 75 seconds.
17
+ - Write the result to `/Users/viraj/Desktop/loadout-demo-1080p.mp4`.
18
+
19
+ ---
20
+
21
+ ### Task 1: Precompute genuine demo artifacts
22
+
23
+ **Files:**
24
+
25
+ - Modify: `/Users/viraj/Desktop/run-loadout-demo.sh`
26
+
27
+ **Interfaces:**
28
+
29
+ - Consumes: current Loadout build, isolated demo state, completed discussion `a71874c7-5ac`.
30
+ - Produces: disposable text outputs and seven Kokoro WAV clips under `.loadout-demo-tmp`.
31
+
32
+ - [ ] **Step 1:** Replace live in-capture commands with preflight commands whose stdout is stored under `.loadout-demo-tmp/output`.
33
+ - [ ] **Step 2:** Generate seven concise `af_heart` narration clips before capture.
34
+ - [ ] **Step 3:** Verify `bash -n /Users/viraj/Desktop/run-loadout-demo.sh` succeeds.
35
+
36
+ ### Task 2: Record a hard-bounded scene timeline
37
+
38
+ **Files:**
39
+
40
+ - Modify: `/Users/viraj/Desktop/run-loadout-demo.sh`
41
+
42
+ **Interfaces:**
43
+
44
+ - Consumes: Task 1 text and WAV artifacts.
45
+ - Produces: `.loadout-demo-tmp/raw.mov` and `.loadout-demo-tmp/narration.wav`.
46
+
47
+ - [ ] **Step 1:** Render seven scenes using saved real output and sleep only for the associated narration duration.
48
+ - [ ] **Step 2:** Use the completed discussion transcript instead of starting paid provider turns.
49
+ - [ ] **Step 3:** Set `screencapture` to a 75-second safety cap and finalize it with `SIGINT`.
50
+
51
+ ### Task 3: Export and verify the launch MP4
52
+
53
+ **Files:**
54
+
55
+ - Modify: `/Users/viraj/Desktop/run-loadout-demo.sh`
56
+ - Create at runtime: `/Users/viraj/Desktop/loadout-demo-1080p.mp4`
57
+
58
+ **Interfaces:**
59
+
60
+ - Consumes: Task 2 video and narration tracks.
61
+ - Produces: one verified H.264/AAC 1920×1080 MP4.
62
+
63
+ - [ ] **Step 1:** Concatenate narration clips with short pauses and mux them with the captured video.
64
+ - [ ] **Step 2:** Scale and pad to exactly 1920×1080 with `yuv420p` and Fast Start.
65
+ - [ ] **Step 3:** Fail if FFprobe reports a duration outside 45–75 seconds or missing video/audio streams.
@@ -0,0 +1,247 @@
1
+ # Maximum + Curator demo recording implementation plan
2
+
3
+ > **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
4
+
5
+ **Goal:** Record, narrate, and export a truthful two-minute Loadout flagship demo to the user’s Desktop.
6
+
7
+ **Architecture:** A disposable TypeScript fixture and a local copy of the published 0.9.4 CLI provide deterministic terminal output. The Maximum install intentionally targets every detected real agent after the approved risk acknowledgement so pre-existing active skills are preserved; the recorded curation and coordination scenes remain focused on Claude Code and Codex. All visible work remains in one clean Terminal window. Screen capture, voice narration, captions, and H.264 export are produced outside the repository.
8
+
9
+ **Tech Stack:** Node.js 20+, npm, Loadout 0.9.4, macOS Terminal, `screencapture`, macOS `say`, FFmpeg, Claude Code CLI, Codex CLI/SDK.
10
+
11
+ ## Global Constraints
12
+
13
+ - Final video duration: 120 seconds or less.
14
+ - Final output: `/Users/viraj/Desktop/loadout-demo-maximum-curator-1080p.mp4`.
15
+ - Capture one terminal rectangle only; do not show any Claude, Codex, browser, Finder, or existing-project UI.
16
+ - Show the real Maximum preview before using `--yes --approve-risk`.
17
+ - Do not enable credentialed MCP integrations.
18
+ - Use real Claude Code and Codex replies for the bounded discussion; label removed idle wait time.
19
+ - Do not commit demo recordings, temporary scripts, audio, or fixture data to this repository.
20
+
21
+ ---
22
+
23
+ ### Task 1: Prepare the isolated recording workspace and published CLI
24
+
25
+ **Files:**
26
+
27
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/package.json`
28
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/fixture/src/api/checkout.ts`
29
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/fixture/src/frontend/checkout.ts`
30
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/fixture/package.json`
31
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/output/raw/`
32
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/output/final/`
33
+
34
+ **Interfaces:**
35
+
36
+ - Consumes: npm registry package `loadout-ai@0.9.4`.
37
+ - Produces: `node_modules/.bin/loadout` resolving to 0.9.4 and a small backend/frontend checkout fixture.
38
+
39
+ - [ ] **Step 1: Create the workspace and fixture files with no personal content.**
40
+
41
+ The fixture API file must export `CheckoutRequest`, `CheckoutResult`, and
42
+ `submitCheckout`. The frontend file must import `CheckoutRequest` from the
43
+ API file, making the contract detector and ownership split meaningful.
44
+
45
+ - [ ] **Step 2: Install exactly the published CLI.**
46
+
47
+ Run:
48
+
49
+ ```bash
50
+ npm install --prefix /Users/viraj/Desktop/.loadout-demo-2026-09-08 loadout-ai@0.9.4
51
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout --version
52
+ ```
53
+
54
+ Expected: `0.9.4`.
55
+
56
+ - [ ] **Step 3: Preflight the recording dependencies.**
57
+
58
+ Run:
59
+
60
+ ```bash
61
+ command -v ffmpeg
62
+ command -v /usr/sbin/screencapture
63
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout recommend --project /Users/viraj/Desktop/.loadout-demo-2026-09-08/fixture
64
+ ```
65
+
66
+ Expected: FFmpeg and macOS screen capture are available; the fixture is
67
+ recognized as a TypeScript/Node project.
68
+
69
+ ### Task 2: Apply Maximum, curate a focused set, and prove rollback
70
+
71
+ **Files:**
72
+
73
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/scripts/prepare-state.sh`
74
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/transcript/setup.txt`
75
+
76
+ **Interfaces:**
77
+
78
+ - Consumes: the published CLI from Task 1 and the user's approved real agent setup.
79
+ - Produces: a verified Maximum library, a curated fixture-specific active set, and rollback evidence.
80
+
81
+ - [ ] **Step 1: Run and retain the Maximum preview.**
82
+
83
+ Run:
84
+
85
+ ```bash
86
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout setup --mode maximum --agents claude-code,codex --api-access none --details
87
+ ```
88
+
89
+ Expected: a summary containing reviewed sources, installable repositories,
90
+ quarantined units, deferred integrations, and the Docker preparation failure.
91
+
92
+ - [ ] **Step 2: Apply the approved Maximum plan to every detected agent.**
93
+
94
+ Run:
95
+
96
+ ```bash
97
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout setup --mode maximum --agents claude-code,codex,cursor,gemini-cli,windsurf --api-access none --yes --approve-risk
98
+ ```
99
+
100
+ Expected: a managed library and rollback snapshot for every detected agent;
101
+ Docker MCP Gateway is explicitly skipped after its preparation failure, and no
102
+ credentialed MCP setup is enabled.
103
+
104
+ - [ ] **Step 3: Curate and apply the fixture working set.**
105
+
106
+ Run:
107
+
108
+ ```bash
109
+ cd /Users/viraj/Desktop/.loadout-demo-2026-09-08/fixture
110
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout recommend --project .
111
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout optimize --project . --agents claude-code,codex --limit 30
112
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout optimize --project . --agents claude-code,codex --limit 30 --yes
113
+ ```
114
+
115
+ Capture concise `status` output after the `--yes` activation transaction.
116
+
117
+ - [ ] **Step 4: Prove reversibility.**
118
+
119
+ Run:
120
+
121
+ ```bash
122
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout rollback --list
123
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout rollback
124
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout status
125
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout optimize --project . --agents claude-code,codex --limit 30 --yes
126
+ ```
127
+
128
+ Expected: the first rollback restores the Maximum-library state that preceded
129
+ the project activation. The final optimize reuses the same deterministic
130
+ project rules so handoff and coordination scenes retain their curated set.
131
+
132
+ - [ ] **Step 5: Verify the real setup remains healthy.**
133
+
134
+ Run:
135
+
136
+ ```bash
137
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout status
138
+ /Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout health
139
+ ```
140
+
141
+ Expected: managed state is readable and no credentialed integration was
142
+ enabled inadvertently.
143
+
144
+ ### Task 3: Capture handoff, coordination, and a real bounded discussion
145
+
146
+ **Files:**
147
+
148
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/scripts/demo-scenes.sh`
149
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/transcript/coordination.txt`
150
+
151
+ **Interfaces:**
152
+
153
+ - Consumes: checkout fixture, Loadout 0.9.4, real Claude Code/Codex credentials.
154
+ - Produces: durable handoff and coordination events including a real three-turn discussion.
155
+
156
+ - [ ] **Step 1: Record a bundled verified handoff.**
157
+
158
+ Set `LOADOUT_BIN` to
159
+ `/Users/viraj/Desktop/.loadout-demo-2026-09-08/node_modules/.bin/loadout` and
160
+ run from the fixture:
161
+
162
+ ```bash
163
+ "$LOADOUT_BIN" handoff codex "Review checkout validation and add edge-case tests" --from claude-code --bundle src/api/checkout.ts --verify "Unit tests cover empty cart IDs"
164
+ "$LOADOUT_BIN" handoff codex
165
+ ```
166
+
167
+ Expected: durable inbox entry with bundle metadata and verification criterion.
168
+
169
+ - [ ] **Step 2: Record ownership and contract coordination.**
170
+
171
+ Run:
172
+
173
+ ```bash
174
+ "$LOADOUT_BIN" coord start --agents claude-code,codex --split backend/frontend
175
+ "$LOADOUT_BIN" coord start --agents claude-code,codex --split backend/frontend --yes
176
+ "$LOADOUT_BIN" coord contract checkout-api --agent claude-code --format typescript --body 'export interface CheckoutAPI { submit(input: CheckoutRequest): Promise<CheckoutResult>; }'
177
+ "$LOADOUT_BIN" coord snapshot codex
178
+ ```
179
+
180
+ Expected: explicit ownership, a TypeScript contract, and a Codex-readable snapshot.
181
+
182
+ - [ ] **Step 3: Record one real three-turn provider exchange.**
183
+
184
+ Run:
185
+
186
+ ```bash
187
+ "$LOADOUT_BIN" coord discuss start "Should checkout validation live in the API handler or domain service?" --agents claude-code,codex --rounds 1 --max-turns 3 --timeout 120 --thread-id demo-checkout-architecture
188
+ "$LOADOUT_BIN" coord replay
189
+ ```
190
+
191
+ Expected: visible provider labels and actual Claude Code/Codex replies, followed
192
+ by a durable replay. If a provider is unavailable, stop the recording and
193
+ report the exact adapter error rather than substituting simulated output.
194
+
195
+ ### Task 4: Record, narrate, and edit the final two-minute video
196
+
197
+ **Files:**
198
+
199
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/audio/narration.txt`
200
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/audio/narration.aiff`
201
+ - Create: `/Users/viraj/Desktop/.loadout-demo-2026-09-08/output/raw/loadout-demo.mov`
202
+ - Create: `/Users/viraj/Desktop/loadout-demo-maximum-curator-1080p.mp4`
203
+
204
+ **Interfaces:**
205
+
206
+ - Consumes: real terminal capture and scene timestamps from Tasks 2–3.
207
+ - Produces: an H.264 1920×1080 video no longer than 120 seconds.
208
+
209
+ - [ ] **Step 1: Prepare the clean terminal framing.**
210
+
211
+ Use one Terminal window, hide unrelated panes, set a large monospace font, and
212
+ size the window to a 960×540-point rectangle. Only that rectangle may be
213
+ recorded.
214
+
215
+ - [ ] **Step 2: Start macOS region capture and run the prepared scene script.**
216
+
217
+ Run:
218
+
219
+ ```bash
220
+ /usr/sbin/screencapture -x -v -V120 -R276,221,960,540 /Users/viraj/Desktop/.loadout-demo-2026-09-08/output/raw/loadout-demo.mov
221
+ ```
222
+
223
+ Capture real scenes, cutting only idle Maximum-install and provider-wait time.
224
+ Add visible cards reading `Maximum install — elapsed time removed` and `Real
225
+ Claude + Codex turns — waiting time removed` at those cuts.
226
+
227
+ - [ ] **Step 3: Generate concise synthetic narration.**
228
+
229
+ Write a 100–120 second narration matching the timed sequence exactly. Generate
230
+ it with an installed macOS synthetic voice, review it for intelligibility, and
231
+ keep it lower than the terminal output during provider turns.
232
+
233
+ - [ ] **Step 4: Export the video.**
234
+
235
+ Run FFmpeg with H.264 video, AAC narration, `yuv420p`, and `+faststart` to
236
+ create `/Users/viraj/Desktop/loadout-demo-maximum-curator-1080p.mp4`.
237
+
238
+ - [ ] **Step 5: Verify the final artifact.**
239
+
240
+ Run:
241
+
242
+ ```bash
243
+ ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 /Users/viraj/Desktop/loadout-demo-maximum-curator-1080p.mp4
244
+ ```
245
+
246
+ Expected: a duration at or below `120.0`. Open the exported file once and
247
+ verify terminal text, captions, narration, and privacy constraints visually.
@@ -0,0 +1,42 @@
1
+ # Fast Launch Demo Design
2
+
3
+ ## Goal
4
+
5
+ Produce a polished 60–75 second Loadout launch video that demonstrates Maximum,
6
+ Optimize/Curator, bundled Handoff, Coordination, and a genuine Claude Code ↔
7
+ Codex Discussion without recording network or provider wait time.
8
+
9
+ ## Story
10
+
11
+ The cut opens with Loadout's one-line value proposition, then moves through five
12
+ proof points: the screened Maximum library, project-aware optimization, a
13
+ context-bundled handoff, explicit ownership/contracts, and a previously completed
14
+ real two-provider discussion. Rollback is omitted to protect pace.
15
+
16
+ ## Capture model
17
+
18
+ Slow read-only commands and demo-state mutations run before screen capture and
19
+ write their real terminal output into a disposable directory. During capture,
20
+ the script displays the exact command followed by that saved output. The video
21
+ labels removed setup or provider waiting time. No agent response is invented.
22
+
23
+ ## Audio and timing
24
+
25
+ Kokoro `af_heart` narrates seven short clips generated before capture. Each scene
26
+ remains visible for the corresponding clip duration, and the clips are assembled
27
+ automatically into the final audio track. The recording is capped at 75 seconds.
28
+
29
+ ## Privacy and output
30
+
31
+ The user records from a fresh full-screen Terminal window. The export is H.264,
32
+ 1920×1080, AAC audio, `yuv420p`, and Fast Start, written to
33
+ `/Users/viraj/Desktop/loadout-demo-1080p.mp4`.
34
+
35
+ ## Acceptance criteria
36
+
37
+ - Final duration is between 45 and 75 seconds.
38
+ - Maximum, Optimize, Handoff, Coordination, and Discussion are visible.
39
+ - Discussion output comes from a genuinely completed Claude Code ↔ Codex thread.
40
+ - No rollback scene, live network wait, paid-provider wait, or unrelated project
41
+ UI appears.
42
+ - The MP4 contains one video stream and one audible AAC audio stream.
@@ -0,0 +1,73 @@
1
+ # Maximum + Curator flagship demo design
2
+
3
+ ## Goal
4
+
5
+ Produce a terminal-only launch demo of no more than two minutes that proves Loadout's complete
6
+ lifecycle without exposing unrelated Claude Code or Codex projects. The video
7
+ uses the user's real agent setup only through terminal commands; it never opens
8
+ either agent's project or conversation UI.
9
+
10
+ ## Audience and promise
11
+
12
+ The video is for developers who have multiple coding agents and too many skills
13
+ to manage. It demonstrates that Loadout can safely assemble a large reviewed
14
+ library, select a project-relevant working set, reverse a change, transfer
15
+ durable context, and let Claude Code and Codex make a bounded decision together.
16
+
17
+ ## Recording environment
18
+
19
+ - Use a newly created, disposable TypeScript checkout fixture with no personal
20
+ files or unrelated repositories.
21
+ - Use the published Loadout 0.9.4 CLI, not the older globally installed binary.
22
+ - Install Maximum into all detected real agent setups only after displaying its
23
+ preview and safety findings and using the user's approved `--yes --approve-risk`
24
+ command. This preserves the user's existing active skills across agents; the
25
+ recorded curation and coordination scenes remain focused on Claude Code and Codex.
26
+ - Keep the capture to a single Terminal window. Do not open Claude Code, Codex,
27
+ a browser, Finder, or any existing project during the take.
28
+ - Capture only the terminal rectangle and use text overlays in post-production.
29
+
30
+ ## Two-minute narrative sequence
31
+
32
+ | Time | Scene | Proof on screen |
33
+ | --------- | ---------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------- |
34
+ | 0:00–0:08 | Hook | “One loadout. Every coding agent.” and the unified workflow image. |
35
+ | 0:08–0:22 | Maximum | Maximum preview summary: reviewed sources, quarantined units, deferred integrations, and the one blocked source. |
36
+ | 0:22–0:38 | Curator | A short time-compressed Maximum apply, then `recommend` and `optimize` selecting a focused TypeScript/Node/Vitest set. |
37
+ | 0:38–0:48 | Rollback | Status, one rollback command, and the restored prior state. |
38
+ | 0:48–1:03 | Handoff | A bundled checkout task with its verification criterion and durable inbox. |
39
+ | 1:03–1:20 | Coordinate | Backend/frontend ownership, a checkout contract, and Codex’s snapshot. |
40
+ | 1:20–1:48 | Discuss for real | One bounded, three-turn Claude Code ↔ Codex discussion. Their provider labels and actual terminal replies are visible; no provider app UI is opened. |
41
+ | 1:48–2:00 | Replay | Durable replay and a concise final payoff card. |
42
+
43
+ ## Safety and truthfulness
44
+
45
+ - The Maximum preview must be visible before applying. State precisely that the
46
+ current plan has quarantined and deferred components; do not imply every
47
+ catalog record is installed.
48
+ - Never configure credentialed MCP services in the demo.
49
+ - Do not show private projects, account identifiers beyond the CLI's ordinary
50
+ provider availability output, secrets, browser tabs, or notification banners.
51
+ - Do not claim a provider exchange is live unless the recorded command actually
52
+ completed with both providers.
53
+ - Maximum writes a managed library and snapshot. The Curator's project-specific
54
+ active set is the recommendation, not an assertion that all Maximum content is
55
+ good for every project.
56
+
57
+ ## Audio and edit
58
+
59
+ - Add a concise synthetic voice walkthrough that explains each proof point in
60
+ plain language. It is narration, not a text-message-song treatment.
61
+ - Use no background track under the narration or provider output; clean terminal
62
+ sound and subtitles take priority.
63
+ - Show real terminal output; compress only inactive install/provider wait time,
64
+ with an explicit on-screen note when time is removed.
65
+
66
+ ## Acceptance criteria
67
+
68
+ - No unrelated project or agent UI is visible in any frame.
69
+ - The final export is 1920×1080 H.264, has a maximum duration of two minutes,
70
+ includes legible terminal text, captions, and synthetic voice narration.
71
+ - The recording shows Maximum, Curator, rollback, a bundled handoff,
72
+ coordination ownership/contracts, a real bounded discussion, and replay.
73
+ - A clean published 0.9.4 CLI is used and the recorded commands succeed.
@@ -0,0 +1,43 @@
1
+ # Loadout Social Launch Copy Design
2
+
3
+ ## Goal
4
+
5
+ Introduce Loadout on X and LinkedIn in Viraj Mishra's established voice: start
6
+ with a concrete problem, explain what he built in plain language, support it
7
+ with specific facts, and end with a direct invitation to try or contribute.
8
+
9
+ ## Positioning
10
+
11
+ Loadout is the package manager for AI coding agent extensions. The launch
12
+ should lead with the fragmented setup problem, then distinguish Loadout through
13
+ project-aware optimization and durable Claude Code/Codex collaboration.
14
+
15
+ Avoid unsupported claims such as "the first ever," "revolutionary," or
16
+ guarantees of virality. Prefer demonstrable facts from the repository and the
17
+ recorded product demo.
18
+
19
+ ## X Structure
20
+
21
+ The main post contains the native video, GitHub link, product hook, and the
22
+ strongest feature summary. It must fit the standard X character limit. A short
23
+ reply thread then explains:
24
+
25
+ 1. Why agent-extension discovery and setup are broken.
26
+ 2. How Maximum and Optimize avoid prompt pollution.
27
+ 3. How Handoff carries source context and verification criteria.
28
+ 4. How Coord uses ownership and contracts for parallel work.
29
+ 5. How Discuss produces a bounded Claude Code/Codex decision.
30
+ 6. The local, preview-first trust model and install command.
31
+
32
+ ## LinkedIn Structure
33
+
34
+ Use one longer post in Viraj's problem → evidence → product → invitation
35
+ pattern. Include concrete repository facts, readable feature bullets, the demo,
36
+ the install command, GitHub link, and a request for feedback, issues, stars, and
37
+ contributions. Do not copy the X thread verbatim.
38
+
39
+ ## Editorial Process
40
+
41
+ Codex writes the fact-checked draft. Claude receives one compact editorial turn
42
+ to challenge the hook, pacing, and authenticity. Codex then produces the final
43
+ paste-ready versions while retaining only claims supported by the repository.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "loadout-ai",
3
- "version": "0.9.3",
3
+ "version": "0.9.5",
4
4
  "private": false,
5
5
  "license": "MIT",
6
6
  "description": "The package manager for AI coding agent extensions",
@@ -46,6 +46,7 @@
46
46
  "node": ">=20"
47
47
  },
48
48
  "scripts": {
49
+ "prepare": "git config core.hooksPath .githooks || true",
49
50
  "clean": "node scripts/clean-dist.mjs",
50
51
  "prebuild": "npm run clean",
51
52
  "build": "tsc -p tsconfig.json",
@@ -86,7 +87,7 @@
86
87
  "devDependencies": {
87
88
  "@eslint/js": "^9.39.5",
88
89
  "@types/node": "^22.10.0",
89
- "@vitest/coverage-v8": "^4.1.10",
90
+ "@vitest/coverage-v8": "^4.1.11",
90
91
  "eslint": "^9.39.5",
91
92
  "eslint-config-prettier": "^10.1.8",
92
93
  "globals": "^17.12.0",
@@ -94,6 +95,6 @@
94
95
  "tsx": "^4.19.2",
95
96
  "typescript": "^5.7.2",
96
97
  "typescript-eslint": "^8.64.0",
97
- "vitest": "^4.1.10"
98
+ "vitest": "^4.1.11"
98
99
  }
99
100
  }