sending ascii bytes to SaaS LLM inference services

llm-slop

129 commits 1 branch 0 tags 445 KiB

Makefile 100%

Find a file

thisago 245df5fba7 docs: readme		2026年07月08日 15:28:53 -03:00
agents	chore(skills,subagents): orgmode, nim. improve	2026年07月06日 13:18:25 +00:00
skills	chore(skills): makefile	2026年07月06日 19:52:30 +00:00
subagents	chore(skills,subagents): orgmode, nim. improve	2026年07月06日 13:18:25 +00:00
.gitignore	refactor: org mode	2026年05月05日 09:13:40 -03:00
.mirrorignore	refactor: split SYSTEM.org, harden subagents	2026年06月10日 09:13:56 -03:00
Makefile	chore: makefile	2026年07月08日 15:28:53 -03:00
README.org	docs: readme	2026年07月08日 15:28:53 -03:00

README.org

Texts usable in LLM applications

Architecture
- Delegation
  - OpenCode
  - Claude Code
Build
Hierarchy

Architecture

Source files are Org Mode. SYSTEM.org contains reusable global rules.

Delegation

Syntax follows the planner delegation pattern, example:

OpenCode

[user] request -> [planner] memory: recall, design ->
-> [user] approval -> [builder] dispatch subagents ->
-> [explorer] software-engineering backend-web: map code paths ->
-> (parallel [developer] backend-web testing: implement
 [developer] backend-web: refactor refs) ->
-> [code-reviewer] software-engineering backend-web: audit ->
-> [developer] backend-web: fix findings ->
-> [builder] evaluate -> [user] feedback

Claude Code

[user] request -> plan mode -> [master] dispatch planner
-> [planner] memory: recall, design -> [master] summarize ->
-> [user] approval -> auto-approve mode -> [master] dispatch subagents ->
-> [explorer] software-engineering backend-web: map code paths ->
-> (parallel [developer] backend-web testing skill3: implement
 [developer] backend-web some-skill2: refactor refs) ->
-> [code-reviewer] software-engineering backend-web: audit ->
-> [developer] backend-web: fix findings ->
-> [master] summarize -> [user] feedback

Build

# export .org -> build/*.md (requires emacs)
make build
# copy build a Claude Code marketplace structure
make mirror MIRROR_TARGET=path/to/plugin
# show all targets
make help

Hierarchy

System prompt: Shared across all agents/subagents.
Agents: Master orchestrators. OpenCode-only concept.
Subagents: Scoped specialists with strict permission boundaries. ClaudeCode + OpenCode.
Skills: Reusable instruction sets loaded by agents/subagents. ClaudeCode + OpenCode.