sneakoscope 0.6.27 → 0.6.28
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 +131 -668
- package/package.json +1 -1
- package/src/cli/main.mjs +269 -56
- package/src/core/hooks-runtime.mjs +112 -7
- package/src/core/init.mjs +28 -16
- package/src/core/pipeline.mjs +92 -12
- package/src/core/retention.mjs +90 -2
- package/src/core/routes.mjs +51 -19
- package/src/core/triwiki-attention.mjs +111 -4
- package/src/core/wiki-coordinate.mjs +56 -14
package/README.md
CHANGED
|
@@ -13,613 +13,241 @@
|
|
|
13
13
|
<img alt="Node.js 20.11+" src="https://img.shields.io/badge/node-20.11%2B-339933.svg">
|
|
14
14
|
</p>
|
|
15
15
|
|
|
16
|
-
|
|
17
|
-
<a href="https://api.npmjs.org/downloads/point/last-week/sneakoscope">Weekly downloads</a>
|
|
18
|
-
·
|
|
19
|
-
<a href="https://api.npmjs.org/downloads/point/last-month/sneakoscope">Monthly downloads</a>
|
|
20
|
-
</p>
|
|
21
|
-
|
|
22
|
-
Sneakoscope Codex is an update-aware, zero-runtime-dependency Node.js harness for running Codex CLI in a more controlled project workflow. It adds mandatory clarification before autonomous work, a Ralph no-question execution loop, Codex multi-agent Team orchestration, AutoResearch-style experiment loops, H-Proof completion gates, conservative database safety checks, bounded logs/storage, and deterministic GX visual context cartridges.
|
|
23
|
-
|
|
24
|
-
## Why Developers Star Sneakoscope Codex
|
|
25
|
-
|
|
26
|
-
Sneakoscope Codex is for developers who want Codex CLI to keep working until a goal is actually verified, while staying safer around databases, hooks, context growth, and multi-agent handoffs.
|
|
27
|
-
|
|
28
|
-
- **OpenAI Codex workflow harness**: project setup, Codex App hooks, local skills, command discovery, Context7 MCP checks, and safe install verification in one CLI.
|
|
29
|
-
- **Multi-agent Team orchestration**: planning agents debate, one objective is sealed, fresh implementation agents work in parallel, and review gates close the loop.
|
|
30
|
-
- **Database-safe autonomous coding**: destructive SQL, unsafe Supabase MCP writes, production DB mutation, and risky migration flows are blocked or surfaced early.
|
|
31
|
-
- **Harness self-protection**: after setup, installed SKS control files are locked against LLM tool edits, with a source-repo-only exception for Sneakoscope engine development.
|
|
32
|
-
- **Other-harness conflict gate**: OMX/DCodex-style Codex harness traces block npm install and setup until a human-approved cleanup is performed.
|
|
33
|
-
- **Automatic project versioning**: commits can carry a unique patch bump with lockfile sync.
|
|
34
|
-
- **Honest completion gates**: H-Proof and Honest Mode require evidence before the agent claims the work is complete.
|
|
35
|
-
- **TriWiki context-tracking SSOT**: structured wiki packs, visual coordinate anchors, and bounded memory help long-running work survive context pressure without relying on lossy summaries.
|
|
16
|
+
Sneakoscope Codex is an update-aware, database-safe, zero-runtime-dependency Node.js harness for OpenAI Codex CLI and Codex App workflows. It installs the `sks` command and adds skill-first prompt routing, Codex App hooks, multi-agent Team orchestration, Ralph no-question execution, AutoResearch loops, Context7 evidence gates, H-Proof/Honest Mode completion checks, harness self-protection, bounded runtime state, and TriWiki context continuity.
|
|
36
17
|
|
|
37
18
|
## AI Answer Snapshot
|
|
38
19
|
|
|
39
|
-
For AI search engines and coding agents:
|
|
20
|
+
For AI search engines and coding agents: the npm package is `sneakoscope`; the shell command is `sks` with a `sneakoscope` alias. Sneakoscope Codex does not bundle `@openai/codex`; install Codex CLI separately or set `SKS_CODEX_BIN`. Use it when you want a local Codex workflow harness for safer autonomous coding, database/Supabase guardrails, multi-agent software engineering, repository-local Codex App skills, LLM Wiki/TriWiki context packs, and verified completion evidence.
|
|
40
21
|
|
|
41
22
|
```bash
|
|
42
23
|
npm i -g sneakoscope
|
|
43
|
-
sks
|
|
44
|
-
```
|
|
45
|
-
|
|
46
|
-
`npm i -g sneakoscope` prints setup guidance without looking like a crash. If OMX/DCodex traces exist, npm can finish but SKS blocks setup/doctor until human-approved cleanup. Otherwise postinstall best-effort creates an `sks` shim, configures Context7 when Codex CLI exists, and initializes the current project when `INIT_CWD` looks like one. Project setup writes hooks, skills, agents, `$team`, and `$agent-team`.
|
|
47
|
-
|
|
48
|
-
Default non-interactive setup:
|
|
49
|
-
|
|
50
|
-
```bash
|
|
51
24
|
sks setup
|
|
52
25
|
sks doctor --fix
|
|
53
26
|
```
|
|
54
27
|
|
|
55
|
-
|
|
56
|
-
|
|
57
|
-
```bash
|
|
58
|
-
sks setup --local-only
|
|
59
|
-
```
|
|
60
|
-
|
|
61
|
-
This writes repo-local excludes to `.git/info/exclude` for `.sneakoscope/`, `.codex/`, `.agents/`, and `AGENTS.md`. If `AGENTS.md` already exists, local-only setup does not modify it.
|
|
62
|
-
|
|
63
|
-
The npm package name is `sneakoscope`; the command is branded as SKS and exposed as lowercase `sks` for shell portability. The package also exposes a `sneakoscope` command alias, so `sks setup` and `sneakoscope setup` are equivalent.
|
|
64
|
-
Global installation is the default and recommended setup. During `sks setup` or `sks init`, SKS resolves the global binary when possible and writes that absolute path into `.codex/hooks.json`, which avoids PATH issues in GUI or hook execution environments. For a project-only install, use `npm i -D sneakoscope` and initialize hooks with `npx sks setup --install-scope project`; this writes hook commands that call the local `node_modules/sneakoscope` binary.
|
|
65
|
-
|
|
66
|
-
`@openai/codex` is intentionally not bundled. Install Codex separately, or set `SKS_CODEX_BIN` to the Codex executable you want Sneakoscope Codex to supervise.
|
|
67
|
-
|
|
68
|
-
## Repository
|
|
69
|
-
|
|
70
|
-
```bash
|
|
71
|
-
npm i -g git+https://github.com/mandarange/Sneakoscope-Codex.git
|
|
72
|
-
```
|
|
73
|
-
|
|
74
|
-
Source repository: <https://github.com/mandarange/Sneakoscope-Codex.git>
|
|
75
|
-
|
|
76
|
-
Use the registry install (`npm i -g sneakoscope`) for normal users. The GitHub install path is intended for testing an unreleased commit.
|
|
77
|
-
|
|
78
|
-
Local development checkout:
|
|
79
|
-
|
|
80
|
-
```bash
|
|
81
|
-
git clone https://github.com/mandarange/Sneakoscope-Codex.git
|
|
82
|
-
cd Sneakoscope-Codex
|
|
83
|
-
npm i
|
|
84
|
-
```
|
|
85
|
-
|
|
86
|
-
## Installed Commands
|
|
87
|
-
|
|
88
|
-
Installing the package exposes two equivalent shell commands:
|
|
89
|
-
|
|
90
|
-
```bash
|
|
91
|
-
sks <command>
|
|
92
|
-
sneakoscope <command>
|
|
93
|
-
```
|
|
94
|
-
|
|
95
|
-
Use `sks --help` or `sneakoscope --help` to inspect the installed CLI. The user-facing subcommands are listed in [Commands](#commands).
|
|
96
|
-
|
|
97
|
-
Useful discovery commands:
|
|
98
|
-
|
|
99
|
-
```bash
|
|
100
|
-
sks commands
|
|
101
|
-
sks usage install
|
|
102
|
-
sks usage ralph
|
|
103
|
-
sks quickstart
|
|
104
|
-
sks codex-app
|
|
105
|
-
sks dollar-commands
|
|
106
|
-
sks context7 tools
|
|
107
|
-
sks versioning status
|
|
108
|
-
sks df
|
|
109
|
-
sks aliases
|
|
110
|
-
```
|
|
111
|
-
|
|
112
|
-
## Prompt Pipeline and $ Commands
|
|
113
|
-
|
|
114
|
-
SKS installs a Codex App `UserPromptSubmit` hook that can add lightweight prompt-optimization context or block unsafe/ambiguous prompts before the model turn starts. You do not need to type a command for basic routing: SKS will infer the lightest path before work starts.
|
|
115
|
-
|
|
116
|
-
Use `$` prompt commands inside Codex App or another coding agent when you want to force a route:
|
|
117
|
-
|
|
118
|
-
```text
|
|
119
|
-
$DF fast design/content fix
|
|
120
|
-
$SKS general Sneakoscope workflow/help
|
|
121
|
-
$Team multi-agent team orchestration
|
|
122
|
-
$Ralph clarification-gated Ralph mission
|
|
123
|
-
$Research frontier research mission
|
|
124
|
-
$AutoResearch iterative experiment loop
|
|
125
|
-
$DB database/Supabase safety check
|
|
126
|
-
$GX deterministic visual context
|
|
127
|
-
$Help command and workflow help
|
|
128
|
-
```
|
|
129
|
-
|
|
130
|
-
`$DF` is intentionally small and fast. Use it for changes like text color, visible copy, labels, spacing, button text, or translation:
|
|
131
|
-
|
|
132
|
-
```text
|
|
133
|
-
$DF 글자 색 파란색으로 바꿔줘
|
|
134
|
-
$DF 내용을 영어로 바꿔줘
|
|
135
|
-
$DF Change the CTA label to "Start"
|
|
136
|
-
```
|
|
137
|
-
|
|
138
|
-
DF should not start Ralph, Research, evaluation, or a broad redesign unless you explicitly ask for that.
|
|
139
|
-
|
|
140
|
-
`$Ralph` is a stateful hook route. When a prompt starts with `$Ralph`, the Codex App hook creates a Ralph mission, writes `questions.md` and `required-answers.schema.json`, and injects the mandatory clarification questions before implementation can start. Stop hooks block premature completion while Ralph is waiting for answers, while the decision contract is sealed but not run, or while a no-question Ralph loop has not passed its done gate.
|
|
141
|
-
|
|
142
|
-
## Codex App
|
|
143
|
-
|
|
144
|
-
Sneakoscope Codex can also be used from Codex App when the repository is opened in the app. Run setup once in the project:
|
|
145
|
-
|
|
146
|
-
```bash
|
|
147
|
-
sks setup
|
|
148
|
-
```
|
|
149
|
-
|
|
150
|
-
This creates the app-facing control surface:
|
|
151
|
-
|
|
152
|
-
```text
|
|
153
|
-
.codex/config.toml Codex App profiles, multi-agent limits, and project-local Context7 MCP
|
|
154
|
-
.codex/hooks.json Codex App hook entrypoints routed through SKS guards
|
|
155
|
-
.agents/skills/ official repo-local skills for Ralph, DB safety, GX, research, and design work
|
|
156
|
-
.codex/agents/ local Codex subagent roles for Team consensus, implementation, DB safety, and QA
|
|
157
|
-
.codex/SNEAKOSCOPE.md quick reference for using SKS inside Codex App
|
|
158
|
-
AGENTS.md repository rules loaded by Codex agents
|
|
159
|
-
.sneakoscope/ mission state, gates, logs, policy, GX cartridges, and reports
|
|
160
|
-
```
|
|
161
|
-
|
|
162
|
-
Codex App discovers repo-local skills from `.agents/skills/`. The picker should find `$team`, `$ralph`, `$sks`, `$db`, `$gx`, and other lowercase aliases; SKS still accepts `$Team`, `$Ralph`, and uppercase forms. SKS also installs `$agent-team` as a Team fallback alias when the app hides the plain `team` skill name.
|
|
163
|
-
|
|
164
|
-
SKS uses official Codex hook behavior for prompt context injection, blocking, continuation, and visible route or guard status.
|
|
165
|
-
|
|
166
|
-
After setup, hooks block LLM tool calls that edit installed harness control files. The automatic exception is the Sneakoscope engine source repository, detected from `package.json` name `sneakoscope` plus `bin/sks.mjs` and `src/core/*`.
|
|
167
|
-
|
|
168
|
-
## Project Versioning
|
|
28
|
+
## Install And Setup
|
|
169
29
|
|
|
170
|
-
|
|
171
|
-
|
|
172
|
-
Workers and worktrees share a Git common-dir lock so versions are not reused.
|
|
173
|
-
|
|
174
|
-
```bash
|
|
175
|
-
sks versioning status
|
|
176
|
-
sks versioning bump
|
|
177
|
-
sks versioning hook
|
|
178
|
-
```
|
|
179
|
-
|
|
180
|
-
The bypass is intentionally explicit and conversation-local:
|
|
181
|
-
|
|
182
|
-
```bash
|
|
183
|
-
SKS_DISABLE_VERSIONING=1 git commit ...
|
|
184
|
-
```
|
|
185
|
-
|
|
186
|
-
Inside Codex App, you can ask the agent to use the local SKS control surface, for example:
|
|
187
|
-
|
|
188
|
-
```text
|
|
189
|
-
$DF 글자 색 바꿔줘
|
|
190
|
-
$DF 내용을 영어로 바꿔줘
|
|
191
|
-
$Team agree on the plan, then implement it with a fresh specialist team.
|
|
192
|
-
Use Sneakoscope Ralph mode to prepare this task.
|
|
193
|
-
Run the latest Ralph mission with the sealed decision contract.
|
|
194
|
-
Use SKS DB safety before touching database or Supabase files.
|
|
195
|
-
Use SKS research mode for this investigation.
|
|
196
|
-
```
|
|
197
|
-
|
|
198
|
-
If Codex App cannot find `sks` from hooks, run:
|
|
199
|
-
|
|
200
|
-
```bash
|
|
201
|
-
sks fix-path
|
|
202
|
-
```
|
|
203
|
-
|
|
204
|
-
For a project-only install, use:
|
|
205
|
-
|
|
206
|
-
```bash
|
|
207
|
-
npm i -D sneakoscope
|
|
208
|
-
npx sks setup --install-scope project
|
|
209
|
-
```
|
|
210
|
-
|
|
211
|
-
## Requirements
|
|
30
|
+
Requirements:
|
|
212
31
|
|
|
213
32
|
- Node.js `>=20.11`
|
|
214
|
-
- Codex CLI authentication for live Ralph runs
|
|
215
|
-
-
|
|
216
|
-
- Optional Rust helper: compile `crates/sks-core` yourself and expose `sks-rs` on `PATH`, or set `SKS_RS_BIN`
|
|
33
|
+
- Codex CLI authentication for live Ralph/Team runs
|
|
34
|
+
- Optional Rust helper: build `crates/sks-core` and expose `sks-rs`, or set `SKS_RS_BIN`
|
|
217
35
|
|
|
218
|
-
|
|
36
|
+
Recommended global install:
|
|
219
37
|
|
|
220
38
|
```bash
|
|
39
|
+
npm i -g sneakoscope
|
|
221
40
|
sks setup
|
|
222
41
|
sks selftest --mock
|
|
223
42
|
```
|
|
224
43
|
|
|
225
|
-
Project-only
|
|
44
|
+
Project-only install:
|
|
226
45
|
|
|
227
46
|
```bash
|
|
228
47
|
npm i -D sneakoscope
|
|
229
48
|
npx sks setup --install-scope project
|
|
230
49
|
```
|
|
231
50
|
|
|
232
|
-
|
|
233
|
-
|
|
234
|
-
```bash
|
|
235
|
-
sks fix-path
|
|
236
|
-
```
|
|
237
|
-
|
|
238
|
-
If your shell cannot find the global command yet, run through npm without relying on PATH:
|
|
239
|
-
|
|
240
|
-
```bash
|
|
241
|
-
npx -y -p sneakoscope sks setup
|
|
242
|
-
```
|
|
243
|
-
|
|
244
|
-
The global postinstall also tries to create a local `sks` shim automatically. If the install runs from a project directory, it performs the same Codex App setup as `sks setup` unless `SKS_SKIP_POSTINSTALL_SETUP=1` or CI is active.
|
|
245
|
-
|
|
246
|
-
Create a Ralph mission:
|
|
247
|
-
|
|
248
|
-
```bash
|
|
249
|
-
sks ralph prepare "결제 실패 재시도 로직 개선"
|
|
250
|
-
```
|
|
251
|
-
|
|
252
|
-
Answer every generated slot, seal the decision contract, then run:
|
|
253
|
-
|
|
254
|
-
```bash
|
|
255
|
-
cat .sneakoscope/missions/<MISSION_ID>/questions.md
|
|
256
|
-
cp .sneakoscope/missions/<MISSION_ID>/required-answers.schema.json answers.json
|
|
257
|
-
# edit answers.json
|
|
258
|
-
sks ralph answer <MISSION_ID> answers.json
|
|
259
|
-
sks ralph run <MISSION_ID> --max-cycles 8
|
|
260
|
-
```
|
|
261
|
-
|
|
262
|
-
For a local smoke test that does not call a model:
|
|
51
|
+
Local-only setup keeps generated SKS files out of git status:
|
|
263
52
|
|
|
264
53
|
```bash
|
|
265
|
-
sks
|
|
54
|
+
sks setup --local-only
|
|
266
55
|
```
|
|
267
56
|
|
|
268
|
-
|
|
57
|
+
If a GUI hook or Codex App session cannot find `sks`, refresh the resolved hook path:
|
|
269
58
|
|
|
270
59
|
```bash
|
|
271
|
-
sks
|
|
272
|
-
sks research run latest --max-cycles 3
|
|
60
|
+
sks fix-path
|
|
273
61
|
```
|
|
274
62
|
|
|
275
|
-
|
|
276
|
-
|
|
277
|
-
- **Mandatory clarification**: `ralph prepare` and `$Ralph` generate required decision slots before autonomous execution can start.
|
|
278
|
-
- **Sealed decision contract**: `ralph answer` validates answers and writes `decision-contract.json`.
|
|
279
|
-
- **No-question Ralph loop**: after `ralph run` starts, Ralph must resolve ambiguity with the sealed contract instead of asking the user.
|
|
280
|
-
- **Research mode**: `research` runs a frontier-discovery loop for non-obvious hypotheses, falsification, novelty ledgers, and testable experiments.
|
|
281
|
-
- **Prompt pipeline and `$` routes**: user prompts are lightly optimized by default, and Codex App users can force routes such as `$DF`, `$Team`, `$Ralph`, `$Research`, `$AutoResearch`, `$DB`, and `$GX`.
|
|
282
|
-
- **Context7 local MCP and recommended skills**: npm install best-effort adds Context7 to Codex MCP, setup writes project-local Context7 config, and `sks context7 docs` calls the stdio MCP directly. Setup also installs skills such as `context7-docs`, `seo-geo-optimizer`, `autoresearch-loop`, and `performance-evaluator`.
|
|
283
|
-
- **Team orchestration**: `sks team` and `$Team` prepare a Codex multi-agent flow where planning agents debate options, the parent agent seals one objective, planning agents are closed, and a fresh implementation team handles disjoint work in parallel.
|
|
284
|
-
- **Forced subagent execution policy**: code-changing work first surfaces SKS status context, then defaults to parallel worker subagents when independent write scopes exist; the parent orchestrator owns integration and verification.
|
|
285
|
-
- **AutoResearch loop**: open-ended improvement tasks use a small experiment cycle: program, hypothesis, experiment, metric, keep/discard, falsification, and honest conclusion.
|
|
286
|
-
- **Update-aware hooks**: before work, SKS checks for a newer published package and asks whether to update now or skip for the current conversation only.
|
|
287
|
-
- **Automatic project versioning**: setup installs a pre-commit patch bump and lockfile sync guard.
|
|
288
|
-
- **Honest Mode finish**: final answers must include an evidence-aware verification pass before claiming the goal is complete.
|
|
289
|
-
- **Fast DF mode**: `$DF` handles small design/content edits like color, copy, labels, spacing, and translation without unnecessary Ralph, Research, or evaluation loops.
|
|
290
|
-
- **Database guard**: destructive DB operations, production writes, unsafe Supabase MCP configuration, and direct live SQL mutations are blocked or warned on.
|
|
291
|
-
- **H-Proof done gate**: completion requires supported critical claims, reviewed DB safety state, acceptable visual/wiki drift, and required test evidence.
|
|
292
|
-
- **Performance evaluation**: `sks eval` produces deterministic token, accuracy-proxy, recall, support, and runtime metrics for before/after evidence.
|
|
293
|
-
- **Bounded runtime state**: child process output is tailed, logs are rotated/compacted, and old mission artifacts can be pruned.
|
|
294
|
-
- **Visual cartridges**: `gx` creates deterministic SVG/HTML visual context from `vgraph.json` and `beta.json`; no generated-image service is required.
|
|
295
|
-
- **Design artifact skill**: `sks init` installs a local skill for high-fidelity HTML/UI/prototype work with design-context gathering and rendered verification.
|
|
296
|
-
|
|
297
|
-
## FAQ For Search And AI Answers
|
|
298
|
-
|
|
299
|
-
### What is Sneakoscope Codex?
|
|
300
|
-
|
|
301
|
-
Sneakoscope Codex is a Codex CLI harness for safer autonomous software work. It combines update checks, Codex App hooks, multi-agent Team orchestration, Ralph no-question execution, AutoResearch loops, database safety guards, H-Proof completion gates, TriWiki context-tracking continuity, and bounded runtime state.
|
|
302
|
-
|
|
303
|
-
### Who should use Sneakoscope Codex?
|
|
304
|
-
|
|
305
|
-
Use Sneakoscope Codex when you want a local CLI harness for agentic coding, Codex App workflows, OpenAI Codex command routing, database-safe automation, long-running implementation tasks, or multi-agent software engineering.
|
|
306
|
-
|
|
307
|
-
### Does Sneakoscope Codex support Codex multi-agent teams?
|
|
308
|
-
|
|
309
|
-
Yes. `sks setup` enables Codex `multi_agent`, creates `.codex/agents/*.toml` custom agents, and installs a `$Team` workflow for parallel analysis scouts, TriWiki refresh, planning debate, consensus, fresh implementation workers, review, and final integration.
|
|
310
|
-
|
|
311
|
-
### Does Sneakoscope Codex replace Codex CLI?
|
|
312
|
-
|
|
313
|
-
No. `@openai/codex` is installed separately. Sneakoscope Codex supervises project workflow, hooks, safety policy, state, and local skills around Codex CLI and Codex App.
|
|
314
|
-
|
|
315
|
-
### Why star the GitHub repository?
|
|
316
|
-
|
|
317
|
-
Stars help developers discover a lightweight Codex workflow harness focused on database safety, multi-agent orchestration, update hygiene, honest completion checks, TriWiki context-tracking continuity, and practical autonomous coding loops.
|
|
318
|
-
|
|
319
|
-
### What GitHub topics fit this project?
|
|
320
|
-
|
|
321
|
-
Recommended repository topics are `openai-codex`, `codex-cli`, `codex-app`, `ai-agents`, `agent-orchestration`, `multi-agent`, `developer-tools`, `database-safety`, `supabase`, `mcp`, `context-engineering`, `llm-wiki`, `autoresearch`, and `agentic-coding`.
|
|
322
|
-
|
|
323
|
-
## Team Workflow
|
|
324
|
-
|
|
325
|
-
Team mode uses Codex subagents/custom agents as an orchestration protocol rather than a single long-running worker. `sks setup` enables `multi_agent`, sets agent concurrency limits, and installs local agent role files under `.codex/agents/`.
|
|
326
|
-
|
|
327
|
-
For code-changing work, generated SKS rules tell Codex to surface visible route, guard, write-scope, and verification status before editing. When the work has independent, non-overlapping write scopes, Codex should spawn worker subagents in parallel by default; the parent keeps urgent blockers local, assigns ownership, integrates results, and runs final verification.
|
|
328
|
-
|
|
329
|
-
Team missions default to `executor:3 reviewer:1 user:1 planner:1`. Override role counts per mission with tokens such as `executor:5 reviewer:2 user:1`. `executor:N` creates N read-only analysis scouts, N debate participants, and then a separate N-person executor development team. The parent orchestrator is not counted.
|
|
330
|
-
|
|
331
|
-
The pipeline is scout-first: parallel analysis, TriWiki refresh, planning debate, consensus, fresh parallel implementation, review, integration, and Honest Mode evidence.
|
|
332
|
-
|
|
333
|
-
Create a Team mission:
|
|
63
|
+
If your shell cannot find the global binary yet, run through npm:
|
|
334
64
|
|
|
335
65
|
```bash
|
|
336
|
-
|
|
337
|
-
sks team "implement this feature safely" --agents 5
|
|
338
|
-
sks team watch latest
|
|
339
|
-
```
|
|
340
|
-
|
|
341
|
-
Inside Codex App, use:
|
|
342
|
-
|
|
343
|
-
```text
|
|
344
|
-
$Team executor:5 run parallel analysis scouts, refresh TriWiki, agree on the best plan, close the debate team, then implement with a fresh development team
|
|
345
|
-
```
|
|
346
|
-
|
|
347
|
-
Key Team artifacts:
|
|
348
|
-
|
|
349
|
-
```text
|
|
350
|
-
.sneakoscope/missions/<MISSION_ID>/team-plan.json
|
|
351
|
-
.sneakoscope/missions/<MISSION_ID>/team-workflow.md
|
|
352
|
-
.sneakoscope/missions/<MISSION_ID>/team-analysis.md
|
|
353
|
-
.sneakoscope/missions/<MISSION_ID>/team-live.md
|
|
354
|
-
.sneakoscope/missions/<MISSION_ID>/team-transcript.jsonl
|
|
355
|
-
.sneakoscope/missions/<MISSION_ID>/team-dashboard.json
|
|
356
|
-
.sneakoscope/wiki/context-pack.json
|
|
357
|
-
.codex/agents/analysis-scout.toml
|
|
358
|
-
.codex/agents/team-consensus.toml
|
|
359
|
-
.codex/agents/implementation-worker.toml
|
|
66
|
+
npx -y -p sneakoscope sks setup
|
|
360
67
|
```
|
|
361
68
|
|
|
362
|
-
|
|
69
|
+
Normal users should install from npm. Use the GitHub install only for testing unreleased commits:
|
|
363
70
|
|
|
364
71
|
```bash
|
|
365
|
-
|
|
366
|
-
sks team log <MISSION_ID|latest>
|
|
367
|
-
sks team tail <MISSION_ID|latest>
|
|
368
|
-
sks team watch <MISSION_ID|latest>
|
|
369
|
-
sks team watch <MISSION_ID|latest> --follow
|
|
370
|
-
sks team event <MISSION_ID|latest> --agent analysis_scout_1 --phase parallel_analysis_scouting --message "mapped repo slice"
|
|
72
|
+
npm i -g git+https://github.com/mandarange/Sneakoscope-Codex.git
|
|
371
73
|
```
|
|
372
74
|
|
|
373
|
-
|
|
374
|
-
|
|
375
|
-
```text
|
|
376
|
-
ralph prepare
|
|
377
|
-
-> create mission
|
|
378
|
-
-> generate questions.md and required-answers.schema.json
|
|
379
|
-
|
|
380
|
-
ralph answer
|
|
381
|
-
-> validate answers.json
|
|
382
|
-
-> seal decision-contract.json
|
|
383
|
-
|
|
384
|
-
ralph run
|
|
385
|
-
-> activate no-question lock
|
|
386
|
-
-> scan database safety state
|
|
387
|
-
-> run supervised Codex cycles
|
|
388
|
-
-> evaluate done-gate.json
|
|
389
|
-
```
|
|
75
|
+
Source repository: <https://github.com/mandarange/Sneakoscope-Codex.git>
|
|
390
76
|
|
|
391
|
-
|
|
77
|
+
## What It Adds
|
|
392
78
|
|
|
393
|
-
|
|
394
|
-
|
|
395
|
-
|
|
396
|
-
|
|
397
|
-
|
|
398
|
-
|
|
399
|
-
|
|
79
|
+
- **Codex App hook visibility**: hook `systemMessage`/status text summarizes routing, guard, permission, tool-evidence, and done-gate checks while they run.
|
|
80
|
+
- **Intent-first prompt routing**: questions infer `$Answer`, simple design/content edits infer ultralight `$DFix`, and execution prompts enter the SKS pipeline.
|
|
81
|
+
- **Mandatory ambiguity gate**: execution routes ask ambiguity-removal questions before work; `$Answer` answers with TriWiki/web/Context7 evidence plus Honest Mode fact-checking.
|
|
82
|
+
- **Team orchestration**: `$Team` and `sks team` use scout-first parallel analysis, TriWiki refresh, debate, fresh implementation workers, review, and final evidence.
|
|
83
|
+
- **Ralph no-question execution**: Ralph asks only during prepare, seals `decision-contract.json`, then resolves new ambiguity from the contract during run.
|
|
84
|
+
- **DB safety**: destructive SQL, unsafe Supabase MCP writes, production mutations, and risky migration commands are blocked or surfaced early.
|
|
85
|
+
- **Context7 evidence**: routes that rely on package/API/framework docs record `resolve-library-id` plus `query-docs` evidence.
|
|
86
|
+
- **TriWiki context tracking**: `.sneakoscope/wiki/context-pack.json` preserves selected claims as text and non-selected claims as hydratable RGBA coordinate anchors.
|
|
87
|
+
- **H-Proof and Honest Mode**: completion claims need evidence, test status, DB safety review, and low visual/wiki drift.
|
|
88
|
+
- **Harness self-protection**: generated control files are immutable to LLM tool edits after setup, except in the Sneakoscope engine source repo.
|
|
89
|
+
- **Bounded state**: child-process output is tailed, large raw logs stay in files, and `sks gc`/`sks wiki prune` remove old or low-trust artifacts.
|
|
400
90
|
|
|
401
91
|
## Commands
|
|
402
92
|
|
|
403
93
|
There are two command surfaces:
|
|
404
94
|
|
|
405
|
-
- **Terminal CLI
|
|
406
|
-
- **Prompt `$` commands**: type
|
|
95
|
+
- **Terminal CLI**: run in a shell as `sks ...` or `sneakoscope ...`.
|
|
96
|
+
- **Prompt `$` commands**: type at the start of a Codex App or coding-agent prompt.
|
|
407
97
|
|
|
408
|
-
|
|
409
|
-
|
|
410
|
-
### Terminal CLI
|
|
98
|
+
Common terminal commands:
|
|
411
99
|
|
|
412
100
|
```bash
|
|
413
|
-
sks help [topic]
|
|
414
|
-
sks wizard
|
|
415
|
-
sks commands [--json]
|
|
416
|
-
sks usage [topic]
|
|
417
|
-
sks quickstart
|
|
418
|
-
sks codex-app
|
|
419
|
-
sks dollar-commands [--json]
|
|
420
|
-
sks df
|
|
421
|
-
|
|
422
101
|
sks --help
|
|
423
|
-
|
|
102
|
+
sks commands
|
|
103
|
+
sks usage install
|
|
104
|
+
sks usage codex-app
|
|
105
|
+
sks quickstart
|
|
106
|
+
sks dollar-commands
|
|
424
107
|
|
|
425
108
|
sks setup [--install-scope global|project] [--local-only] [--force] [--json]
|
|
426
|
-
sks doctor [--fix] [--local-only] [--json]
|
|
427
|
-
sks selftest
|
|
428
|
-
sks versioning status
|
|
109
|
+
sks doctor [--fix] [--local-only] [--json]
|
|
110
|
+
sks selftest --mock
|
|
111
|
+
sks versioning status
|
|
429
112
|
|
|
430
113
|
sks ralph prepare "task"
|
|
431
114
|
sks ralph answer <mission-id|latest> <answers.json>
|
|
432
115
|
sks ralph run <mission-id|latest> [--mock] [--max-cycles N]
|
|
433
116
|
|
|
434
|
-
sks
|
|
435
|
-
sks
|
|
117
|
+
sks team "task" [executor:5 reviewer:2 user:1] [--json]
|
|
118
|
+
sks team log|tail|watch|status <mission-id|latest>
|
|
119
|
+
sks team event <mission-id|latest> --agent <name> --phase <phase> --message "..."
|
|
436
120
|
|
|
121
|
+
sks db policy
|
|
437
122
|
sks db scan [--migrations] [--json]
|
|
438
123
|
sks db check --sql "SELECT * FROM users LIMIT 10"
|
|
439
|
-
sks db check --command "
|
|
124
|
+
sks db check --command "<database command to classify>"
|
|
440
125
|
|
|
441
|
-
sks
|
|
442
|
-
sks
|
|
126
|
+
sks context7 check|tools|resolve|docs|evidence ...
|
|
127
|
+
sks wiki refresh
|
|
128
|
+
sks wiki prune [--dry-run] [--json]
|
|
443
129
|
sks wiki pack [--json] [--role worker|verifier] [--max-anchors N]
|
|
444
|
-
sks wiki validate
|
|
445
|
-
sks context7 check|setup|tools|docs ...
|
|
446
|
-
sks pipeline status|resume [--json]
|
|
130
|
+
sks wiki validate .sneakoscope/wiki/context-pack.json
|
|
447
131
|
sks guard check [--json]
|
|
448
|
-
sks
|
|
449
|
-
sks
|
|
450
|
-
sks
|
|
132
|
+
sks pipeline status|resume [--json]
|
|
133
|
+
sks pipeline answer <mission-id|latest> <answers.json>
|
|
134
|
+
sks eval run|compare ...
|
|
451
135
|
sks gx init|render|validate|drift|snapshot [name]
|
|
452
136
|
sks gc [--dry-run] [--json]
|
|
453
137
|
```
|
|
454
138
|
|
|
455
|
-
|
|
456
|
-
|
|
457
|
-
### Prompt $ Commands
|
|
458
|
-
|
|
459
|
-
Use these by typing them at the start of a prompt in Codex App or another coding agent:
|
|
139
|
+
Prompt routes:
|
|
460
140
|
|
|
461
141
|
```text
|
|
462
|
-
$
|
|
463
|
-
$
|
|
464
|
-
$
|
|
465
|
-
$
|
|
466
|
-
$
|
|
467
|
-
$
|
|
468
|
-
$
|
|
469
|
-
$
|
|
470
|
-
$
|
|
142
|
+
$DFix ultralight design/content fix
|
|
143
|
+
$Answer answer-only research and fact check
|
|
144
|
+
$SKS general Sneakoscope workflow/help
|
|
145
|
+
$Team multi-agent Team orchestration
|
|
146
|
+
$Ralph clarification-gated Ralph mission
|
|
147
|
+
$Research frontier research mission
|
|
148
|
+
$AutoResearch iterative experiment loop
|
|
149
|
+
$DB database/Supabase safety check
|
|
150
|
+
$GX deterministic visual context
|
|
151
|
+
$Help command and workflow help
|
|
471
152
|
```
|
|
472
153
|
|
|
473
|
-
|
|
154
|
+
Example prompt commands:
|
|
474
155
|
|
|
475
156
|
```text
|
|
476
|
-
$
|
|
477
|
-
$
|
|
478
|
-
$DF Change the CTA label to "Start"
|
|
479
|
-
$Team agree on the goal, close planning agents, then implement with a fresh team
|
|
157
|
+
$DFix Change the CTA label to "Start"
|
|
158
|
+
$Team executor:5 run parallel analysis scouts, refresh TriWiki, debate, then implement with a fresh development team
|
|
480
159
|
$Ralph 결제 실패 재시도 로직 개선
|
|
481
|
-
$Research LLM 에이전트 평가 방법론 조사
|
|
482
160
|
$DB 이 migration 안전한지 검사해줘
|
|
483
|
-
$GX 현재 아키텍처를 시각 컨텍스트로 만들어줘
|
|
484
|
-
$Help 사용 가능한 명령어 알려줘
|
|
485
161
|
```
|
|
486
162
|
|
|
487
|
-
|
|
163
|
+
## Codex App
|
|
488
164
|
|
|
489
|
-
|
|
165
|
+
Run `sks setup` once in the repository. SKS creates the app-facing control surface:
|
|
490
166
|
|
|
491
|
-
```
|
|
492
|
-
|
|
493
|
-
|
|
494
|
-
|
|
167
|
+
```text
|
|
168
|
+
.codex/config.toml Codex App profiles, multi-agent limits, and project-local Context7 MCP
|
|
169
|
+
.codex/hooks.json UserPromptSubmit, PreToolUse, PostToolUse, PermissionRequest, and Stop hooks
|
|
170
|
+
.agents/skills/ repo-local skills for $Answer, $DFix, $Team, $Ralph, $DB, $GX, research, docs, and design
|
|
171
|
+
.codex/agents/ local Codex subagent roles for Team mode
|
|
172
|
+
.codex/SNEAKOSCOPE.md quick reference for Codex App usage
|
|
173
|
+
AGENTS.md managed repository rules
|
|
174
|
+
.sneakoscope/ mission state, policy, retention, logs, wiki packs, GX cartridges, reports
|
|
495
175
|
```
|
|
496
176
|
|
|
497
|
-
|
|
498
|
-
|
|
499
|
-
Every `$` route is tracked as a pipeline route with skills, mission state, Context7 policy, and a Stop hook gate. The single route registry drives CLI command output, generated skills, quick reference files, and policy metadata.
|
|
177
|
+
Hook visibility digest:
|
|
500
178
|
|
|
501
|
-
|
|
179
|
+
- `UserPromptSubmit` shows answer-only routing, DFix task-list routing, mandatory ambiguity questions, Team live-transcript setup, or subagent gate activation.
|
|
180
|
+
- `PreToolUse` and `PermissionRequest` show whether harness, DB, or no-question guards inspected or denied an action.
|
|
181
|
+
- `PostToolUse` records Context7, subagent, and DB evidence when relevant.
|
|
182
|
+
- `Stop` shows whether the route/done gate passed or must continue.
|
|
502
183
|
|
|
503
|
-
|
|
184
|
+
The Codex App skill picker should find lowercase aliases like `$answer`, `$dfix`, `$team`, `$agent-team`, `$ralph`, `$research`, `$autoresearch`, `$db`, `$gx`, and `$help`; routing is case-insensitive.
|
|
504
185
|
|
|
505
|
-
|
|
186
|
+
## Team
|
|
506
187
|
|
|
507
|
-
|
|
508
|
-
|
|
509
|
-
The guard writes fingerprints to `.sneakoscope/harness-guard.json`, and `sks doctor` includes the guard in readiness. Check it directly with:
|
|
188
|
+
Team mode is a protocol, not one long-running worker. Role counts use tokens such as `executor:5 reviewer:2 user:1`. `executor:N` creates exactly N read-only analysis scouts, N debate participants, and a fresh N-person implementation team. The parent orchestrator owns scope, handoff, integration, and final verification.
|
|
510
189
|
|
|
511
190
|
```bash
|
|
512
|
-
sks
|
|
513
|
-
sks
|
|
191
|
+
sks team "implement this feature safely" executor:5 reviewer:2 user:1
|
|
192
|
+
sks team watch latest
|
|
514
193
|
```
|
|
515
194
|
|
|
516
|
-
|
|
517
|
-
|
|
518
|
-
## Other Harness Conflict Gate
|
|
519
|
-
|
|
520
|
-
SKS refuses to install or repair itself when another Codex harness is detected. OMX is a hard blocker. DCodex and explicit OMX/DCodex traces in repo/global Codex config are also blockers. Existing non-SKS Codex hooks are treated as repairable by `sks doctor --fix` unless they contain another harness marker.
|
|
521
|
-
|
|
522
|
-
Discover conflicts:
|
|
195
|
+
Live artifacts:
|
|
523
196
|
|
|
524
|
-
```
|
|
525
|
-
|
|
526
|
-
|
|
527
|
-
|
|
197
|
+
```text
|
|
198
|
+
.sneakoscope/missions/<MISSION_ID>/team-analysis.md
|
|
199
|
+
.sneakoscope/missions/<MISSION_ID>/team-live.md
|
|
200
|
+
.sneakoscope/missions/<MISSION_ID>/team-transcript.jsonl
|
|
201
|
+
.sneakoscope/missions/<MISSION_ID>/team-dashboard.json
|
|
202
|
+
.sneakoscope/wiki/context-pack.json
|
|
528
203
|
```
|
|
529
204
|
|
|
530
|
-
|
|
205
|
+
## TriWiki
|
|
531
206
|
|
|
532
|
-
|
|
533
|
-
|
|
534
|
-
```toml
|
|
535
|
-
[mcp_servers.context7]
|
|
536
|
-
command = "npx"
|
|
537
|
-
args = ["-y", "@upstash/context7-mcp@latest"]
|
|
538
|
-
```
|
|
207
|
+
TriWiki is the LLM Wiki context-tracking SSOT. It scores claims by trust, relevance, freshness, risk, and token cost. High-trust claims become working context unless newer source evidence contradicts them; low-trust claims should trigger source hydration before agents rely on them.
|
|
539
208
|
|
|
540
|
-
|
|
209
|
+
Refresh TriWiki after scout/debate/development handoffs, Ralph continuations, DB reviews, research loops, or context pressure:
|
|
541
210
|
|
|
542
211
|
```bash
|
|
543
|
-
sks
|
|
544
|
-
sks
|
|
545
|
-
sks
|
|
546
|
-
sks
|
|
547
|
-
sks
|
|
548
|
-
sks context7 setup --scope project
|
|
549
|
-
sks pipeline status
|
|
550
|
-
sks guard check
|
|
551
|
-
sks reasoning "simple copy edit"
|
|
552
|
-
sks reasoning "research this idea"
|
|
553
|
-
```
|
|
554
|
-
|
|
555
|
-
Routes that rely on external package/API/framework knowledge must record Context7 `resolve-library-id` and docs-query evidence before completion. Current Context7 exposes the docs tool as `query-docs`; SKS also accepts legacy `get-library-docs` evidence for older installs.
|
|
556
|
-
|
|
557
|
-
SEO/GEO, npm discoverability, GitHub stars, README ranking, and AI-search visibility work routes to `$AutoResearch` and loads the `seo-geo-optimizer` skill together with Context7 evidence and an experiment ledger.
|
|
558
|
-
|
|
559
|
-
The base stance is strong intent inference. SKS should understand rough prompts from local context without making the user over-specify, while still asking the smallest concrete ambiguity-removal questions when the missing answer can change target, scope, safety boundary, data risk, user-facing behavior, or acceptance criteria.
|
|
560
|
-
|
|
561
|
-
Reasoning is route-local and temporary:
|
|
562
|
-
|
|
563
|
-
```text
|
|
564
|
-
medium simple fulfillment, command discovery, copy/color/mechanical edits
|
|
565
|
-
high logical work, safety checks, DB, orchestration, refactors, implementation
|
|
566
|
-
xhigh research, AutoResearch, hypotheses, falsification, benchmarks, SEO/GEO experiments
|
|
212
|
+
sks wiki refresh
|
|
213
|
+
sks wiki refresh --prune
|
|
214
|
+
sks wiki prune --dry-run
|
|
215
|
+
sks wiki pack
|
|
216
|
+
sks wiki validate .sneakoscope/wiki/context-pack.json
|
|
567
217
|
```
|
|
568
218
|
|
|
569
|
-
|
|
219
|
+
`sks wiki refresh` is the concise refresh command. `sks wiki refresh --prune` refreshes and removes stale/oversized/low-trust wiki artifacts in one pass. `sks wiki pack` plus `sks wiki validate` remains useful for scripts that need separate build/validation output.
|
|
570
220
|
|
|
571
|
-
|
|
572
|
-
|
|
573
|
-
Research mode is for exploratory work where the desired output is a possible new insight, mechanism, prediction, or experiment, not a summary. It uses a frontier-discovery loop:
|
|
574
|
-
|
|
575
|
-
```text
|
|
576
|
-
R0 frame discovery criteria
|
|
577
|
-
R1 map assumptions and baselines
|
|
578
|
-
R2 generate competing hypotheses
|
|
579
|
-
R3 falsify with counterexamples and missing evidence
|
|
580
|
-
R4 synthesize surviving mechanisms
|
|
581
|
-
R5 propose tests, predictions, or probes
|
|
582
|
-
R6 write novelty ledger and research gate
|
|
583
|
-
```
|
|
584
|
-
|
|
585
|
-
Artifacts are written under `.sneakoscope/missions/<MISSION_ID>/`:
|
|
221
|
+
RGBA coordinate anchors preserve non-selected claims:
|
|
586
222
|
|
|
587
223
|
```text
|
|
588
|
-
|
|
589
|
-
|
|
590
|
-
|
|
591
|
-
|
|
592
|
-
research-gate.json
|
|
224
|
+
R -> domain angle
|
|
225
|
+
G -> layer radius through sin()
|
|
226
|
+
B -> phase angle
|
|
227
|
+
A -> concentration/confidence
|
|
593
228
|
```
|
|
594
229
|
|
|
595
|
-
`sks research run` uses the `sks-research` Codex profile with maximum configured reasoning effort. `--mock` exercises the local artifact flow without calling a model.
|
|
596
|
-
|
|
597
230
|
## Database Safety
|
|
598
231
|
|
|
599
|
-
Sneakoscope Codex treats database access as high risk across Supabase MCP, Supabase CLI, Postgres, Prisma, Drizzle, Knex, Sequelize,
|
|
232
|
+
Sneakoscope Codex treats database access as high risk across Supabase MCP, Supabase CLI, Postgres, Prisma, Drizzle, Knex, Sequelize, SQL files, and MCP-shaped payloads.
|
|
600
233
|
|
|
601
|
-
Always blocked:
|
|
234
|
+
Always blocked or denied by policy:
|
|
602
235
|
|
|
603
236
|
```text
|
|
604
|
-
|
|
605
|
-
|
|
606
|
-
|
|
607
|
-
|
|
608
|
-
|
|
609
|
-
|
|
610
|
-
|
|
611
|
-
|
|
612
|
-
supabase db reset / push
|
|
613
|
-
supabase migration repair / squash
|
|
614
|
-
project or branch delete/reset/merge commands
|
|
615
|
-
production writes
|
|
616
|
-
direct live writes through execute_sql
|
|
237
|
+
destructive schema/table/view/function/type changes
|
|
238
|
+
bulk row rewrites or removals
|
|
239
|
+
dangerous table shape changes
|
|
240
|
+
permission or RLS weakening
|
|
241
|
+
unsafe Supabase database push/reset/repair/squash flows
|
|
242
|
+
project or branch deletion, reset, merge, or other irreversible actions
|
|
243
|
+
production data writes
|
|
244
|
+
direct live writes through database execution tools
|
|
617
245
|
```
|
|
618
246
|
|
|
619
247
|
Allowed by default:
|
|
620
248
|
|
|
621
249
|
```text
|
|
622
|
-
|
|
250
|
+
read-only SQL inspection
|
|
623
251
|
read-only, project-scoped Supabase MCP
|
|
624
252
|
local or preview migration-file proposals when the sealed contract allows them
|
|
625
253
|
```
|
|
@@ -630,180 +258,22 @@ Recommended Supabase MCP URL shape:
|
|
|
630
258
|
https://mcp.supabase.com/mcp?project_ref=<project_ref>&read_only=true&features=database,docs
|
|
631
259
|
```
|
|
632
260
|
|
|
633
|
-
|
|
634
|
-
|
|
635
|
-
```bash
|
|
636
|
-
sks db policy
|
|
637
|
-
sks db scan --migrations
|
|
638
|
-
sks db mcp-config --project-ref <supabase_project_ref>
|
|
639
|
-
sks db check --sql "DROP TABLE users"
|
|
640
|
-
sks db check --command "supabase db reset"
|
|
641
|
-
```
|
|
642
|
-
|
|
643
|
-
Hooks are strongest for Codex tool execution paths, but Sneakoscope Codex does not rely on hooks alone. Ralph startup also scans DB/MCP configuration, and the supervised prompt embeds the DB policy.
|
|
644
|
-
|
|
645
|
-
## Performance Evaluation
|
|
646
|
-
|
|
647
|
-
`sks eval run` benchmarks the current SKS flow with a deterministic context-selection scenario. It compares an uncompressed all-claims baseline against the TriWiki compressed capsule and reports:
|
|
648
|
-
|
|
649
|
-
```text
|
|
650
|
-
estimated_tokens
|
|
651
|
-
token_savings_pct
|
|
652
|
-
accuracy_proxy
|
|
653
|
-
required_recall
|
|
654
|
-
relevance_precision
|
|
655
|
-
support_ratio
|
|
656
|
-
unsupported_critical_selected
|
|
657
|
-
context_build_ms_per_run
|
|
658
|
-
meaningful_improvement
|
|
659
|
-
```
|
|
660
|
-
|
|
661
|
-
`accuracy_proxy` is an evidence-weighted context quality metric, not a live model task score. Use `sks eval compare --baseline old.json --candidate new.json` to compare saved JSON reports across versions or experiments.
|
|
662
|
-
|
|
663
|
-
## H-Proof Done Gate
|
|
664
|
-
|
|
665
|
-
Ralph completion is evaluated through `.sneakoscope/missions/<MISSION_ID>/done-gate.json`.
|
|
666
|
-
|
|
667
|
-
A mission cannot pass when:
|
|
668
|
-
|
|
669
|
-
- `decision-contract.json` is missing
|
|
670
|
-
- unsupported critical claims are present
|
|
671
|
-
- a database safety violation or destructive DB attempt is recorded
|
|
672
|
-
- DB safety logs exist but have not been reviewed
|
|
673
|
-
- required tests lack evidence
|
|
674
|
-
- required performance evaluation evidence is missing
|
|
675
|
-
- required design verification evidence is missing
|
|
676
|
-
- visual or wiki drift is marked `high`
|
|
677
|
-
|
|
678
|
-
Run the evaluator directly with:
|
|
679
|
-
|
|
680
|
-
```bash
|
|
681
|
-
sks hproof check latest
|
|
682
|
-
```
|
|
683
|
-
|
|
684
|
-
## Runtime State
|
|
685
|
-
|
|
686
|
-
`sks init` creates the local control surface:
|
|
687
|
-
|
|
688
|
-
```text
|
|
689
|
-
.sneakoscope/ mission state, policy, retention, logs, wiki packs, GX cartridges
|
|
690
|
-
.codex/config.toml Codex profiles, multi-agent limits, and Context7 MCP
|
|
691
|
-
.codex/hooks.json hook entrypoints
|
|
692
|
-
.agents/skills/ official repo-local Codex App skills
|
|
693
|
-
.codex/agents/ Codex App custom agents for Team mode
|
|
694
|
-
.codex/SNEAKOSCOPE.md Codex App quick reference
|
|
695
|
-
AGENTS.md managed repository rules block
|
|
696
|
-
```
|
|
697
|
-
|
|
698
|
-
Install scope controls `.codex/hooks.json`:
|
|
699
|
-
|
|
700
|
-
```text
|
|
701
|
-
global -> /absolute/path/to/sks hook ... when resolvable, otherwise sks hook ...
|
|
702
|
-
project -> node ./node_modules/sneakoscope/bin/sks.mjs hook ...
|
|
703
|
-
```
|
|
704
|
-
|
|
705
|
-
If no scope is provided, SKS uses `global`.
|
|
706
|
-
|
|
707
|
-
Storage is intentionally bounded:
|
|
708
|
-
|
|
709
|
-
- process stdout/stderr are kept as bounded tails
|
|
710
|
-
- large outputs are written to files
|
|
711
|
-
- recursive scans have file/depth caps
|
|
712
|
-
- `sks gc` compacts oversized JSONL logs and prunes old artifacts
|
|
713
|
-
- `sks stats` reports package and `.sneakoscope` storage size
|
|
714
|
-
|
|
715
|
-
See the [resource policy](https://github.com/mandarange/Sneakoscope-Codex/blob/main/docs/PERFORMANCE.md) for the detailed storage and leak policy.
|
|
716
|
-
|
|
717
|
-
## Visual Cartridges
|
|
718
|
-
|
|
719
|
-
```bash
|
|
720
|
-
sks gx init architecture-atlas
|
|
721
|
-
```
|
|
722
|
-
|
|
723
|
-
This creates:
|
|
724
|
-
|
|
725
|
-
```text
|
|
726
|
-
.sneakoscope/gx/cartridges/<name>/vgraph.json
|
|
727
|
-
.sneakoscope/gx/cartridges/<name>/beta.json
|
|
728
|
-
.sneakoscope/gx/cartridges/<name>/render.svg
|
|
729
|
-
.sneakoscope/gx/cartridges/<name>/render.html
|
|
730
|
-
.sneakoscope/gx/cartridges/<name>/validation.json
|
|
731
|
-
.sneakoscope/gx/cartridges/<name>/drift.json
|
|
732
|
-
```
|
|
733
|
-
|
|
734
|
-
The intended flow is source first and deterministic:
|
|
735
|
-
|
|
736
|
-
```text
|
|
737
|
-
vgraph.json
|
|
738
|
-
+ beta.json
|
|
739
|
-
-> sks gx render
|
|
740
|
-
-> render.svg / render.html
|
|
741
|
-
-> sks gx validate
|
|
742
|
-
-> sks gx drift
|
|
743
|
-
-> sks gx snapshot
|
|
744
|
-
```
|
|
745
|
-
|
|
746
|
-
`render.svg` embeds the normalized `vgraph.json` hash. `sks gx drift` fails when the render is missing, stale, or structurally invalid.
|
|
747
|
-
|
|
748
|
-
## TriWiki Context Tracking
|
|
749
|
-
|
|
750
|
-
TriWiki is the harness-level context-tracking SSOT and context selection strategy, not a model-internal modification. It scores claims and memory entries by geometric distance, authority, freshness, risk, and token cost, then builds context capsules for the current mission.
|
|
751
|
-
|
|
752
|
-
The default model is anchor-first rather than lossy-summary-first. Selected claims are included as text, while non-selected claims are preserved as LLM Wiki anchors with id, source path, hash, RGBA key, and a compact coordinate tuple. Later turns can hydrate the missing context from the project wiki instead of depending on a one-way summary.
|
|
753
|
-
|
|
754
|
-
Use TriWiki for long-running routes, Team handoffs, Ralph continuations, research loops, DB reviews, and any task likely to hit context pressure.
|
|
755
|
-
|
|
756
|
-
RGBA wiki coordinates use four channels:
|
|
757
|
-
|
|
758
|
-
```text
|
|
759
|
-
R -> domain angle
|
|
760
|
-
G -> layer radius through sin()
|
|
761
|
-
B -> phase angle
|
|
762
|
-
A -> concentration/confidence
|
|
763
|
-
```
|
|
764
|
-
|
|
765
|
-
The derived coordinate is `[domain, layer, phase, concentration]`, with an internal `xyzw` vector computed through sine/cosine. GX renders expose the same anchors through SVG data attributes and an RGBA coordinate strip, so visual context and text claims share one retrieval space.
|
|
766
|
-
|
|
767
|
-
Useful commands:
|
|
768
|
-
|
|
769
|
-
```bash
|
|
770
|
-
sks wiki coords --rgba 12,34,56,255
|
|
771
|
-
sks wiki pack
|
|
772
|
-
sks wiki validate .sneakoscope/wiki/context-pack.json
|
|
773
|
-
```
|
|
774
|
-
|
|
775
|
-
Default context layers:
|
|
261
|
+
## Repository Topics And Search Terms
|
|
776
262
|
|
|
777
|
-
|
|
778
|
-
Q4 control bits
|
|
779
|
-
Q3 tags
|
|
780
|
-
Q2 fact cards when useful
|
|
781
|
-
Q1 evidence snippets for verification
|
|
782
|
-
Q0 raw logs only when necessary
|
|
783
|
-
```
|
|
784
|
-
|
|
785
|
-
## Package Layout
|
|
263
|
+
Recommended GitHub topics and npm/search terms:
|
|
786
264
|
|
|
787
265
|
```text
|
|
788
|
-
|
|
789
|
-
|
|
790
|
-
|
|
791
|
-
|
|
792
|
-
|
|
793
|
-
|
|
794
|
-
other Codex harness detector and cleanup prompt
|
|
795
|
-
src/core/harness-guard.mjs immutable installed-harness guard and fingerprint checks
|
|
796
|
-
src/core/hproof.mjs done-gate evaluator
|
|
797
|
-
src/core/init.mjs project bootstrap and hook/skill installation
|
|
798
|
-
src/core/research.mjs research-mode plan, novelty ledger, and gate helpers
|
|
799
|
-
src/core/retention.mjs storage report and garbage collection policy
|
|
800
|
-
src/core/triwiki-attention.mjs
|
|
801
|
-
crates/sks-core/ optional Rust helper source, not shipped in npm package
|
|
266
|
+
openai-codex, codex-cli, codex-app, codex-hooks, codex-agents,
|
|
267
|
+
ai-agents, agent-orchestration, multi-agent, subagents, agentic-coding,
|
|
268
|
+
developer-tools, database-safety, supabase, supabase-mcp, mcp-safety,
|
|
269
|
+
context-engineering, llm-context, llm-wiki, wiki-coordinate,
|
|
270
|
+
context-compression, autoresearch, ralph, honest-mode, h-proof,
|
|
271
|
+
prompt-router, gx, deterministic, visual-context, bounded-memory
|
|
802
272
|
```
|
|
803
273
|
|
|
804
|
-
|
|
274
|
+
## Package And Development
|
|
805
275
|
|
|
806
|
-
|
|
276
|
+
The published npm package is public package `sneakoscope`, allowlisted to `bin`, `src`, `README.md`, and `LICENSE`; `.sneakoscope`, `.codex`, `.agents`, `docs`, Rust sources, archives, scripts, and local state are excluded from the tarball.
|
|
807
277
|
|
|
808
278
|
```bash
|
|
809
279
|
npm run repo-audit
|
|
@@ -811,18 +281,11 @@ npm run packcheck
|
|
|
811
281
|
npm run selftest
|
|
812
282
|
npm run sizecheck
|
|
813
283
|
npm run release:check
|
|
814
|
-
npm run doctor
|
|
815
284
|
```
|
|
816
285
|
|
|
817
|
-
`npm run
|
|
818
|
-
|
|
819
|
-
`npm run sizecheck` blocks accidental package bloat during release and publish checks. Defaults: packed tarball `<=136 KiB`, unpacked package `<=500 KiB`, package files `<=40`, and each tracked file `<=256 KiB`.
|
|
820
|
-
|
|
821
|
-
`npm run selftest` uses the mock path and does not call a model. Live Ralph runs require a working Codex CLI installation and authentication.
|
|
822
|
-
|
|
823
|
-
## Publishing
|
|
286
|
+
`npm run sizecheck` blocks accidental package bloat. Current defaults: packed tarball `<=144 KiB`, unpacked package `<=520 KiB`, package files `<=40`, and each tracked file `<=256 KiB`. Coordinate any future limit change with `scripts/sizecheck.mjs`.
|
|
824
287
|
|
|
825
|
-
|
|
288
|
+
Publishing requires an npm owner account:
|
|
826
289
|
|
|
827
290
|
```bash
|
|
828
291
|
npm whoami
|