A context operating system for humans and coding agents
Manifest-first · two vaults · fail-closed AppSec · S/M/C effort · PT · EN · ES · 中文

---

Agent Context Pack

Most agent setups fail the same way: prompts live in chat memory, rules drift, secrets leak into “helpful” examples, and every new model relearns your workflow from scratch.

Agent Context Pack is a reusable, public template for a personal (or team) context OS:

1. One Manifest agents must read before non-trivial work
2. Hard agent rules (no invention, tools over memory, fail-closed security)
3. Working style with S / M / C effort classes
4. AppSec playbook you can copy into audits
5. Identity / brand / stack skeletons with placeholders
6. UX criteria for agent-built UI
7. Repo boot templates (CLAUDE.md, maps, decisions)

This repo ships scrubbed content only. Real names, firms, products, emails, and paths are replaced by tokens like [YOUR_NAME] and [DOMAIN_VAULT]. See PLACEHOLDERS.md. Keep a private overlay for your real board, vaults, and secrets.

Why this shape

Problem	What this pack does
Agents invent when evidence is missing	Fail-closed rules: mark UNVERIFIED, never fill gaps
Context is scattered across chats	Canonical files under docs/<lang>/ beat chat memory
Domain knowledge gets improvised	Two-vault routing: ops vs domain second brain
Tiny edits and prod auth get the same ceremony	S / M / C effort classes match blast radius
Public templates leak private life	Placeholders + leak-scan posture before publish

Choose your language

	Language	Start here
🇧🇷	Português	docs/pt/README.md
🇺🇸	English	docs/en/README.md
🇪🇸	Español	docs/es/README.md
🇨🇳	中文	docs/zh/README.md

Each language folder contains the full Camada 1 (not summaries): Manifest, rules, working-style, AppSec, identity, brand, stack, implementer DoD, UX pack, and templates.

Mental model (60 seconds)

                 ┌─────────────────────┐
                 │   MANIFEST.md       │  read first
                 └─────────┬───────────┘
                           │
         ┌─────────────────┼─────────────────┐
         ▼                 ▼                 ▼
  agent_rules.md    working-style.md   appsec-rules.md
  (hard rules)      (S/M/C + collab)    (fail-closed)
         │                 │                 │
         └────────┬────────┴────────┬────────┘
                  ▼                 ▼
           [OPS_VAULT]        [DOMAIN_VAULT]
        capture→wiki→MOCs    domain second brain

Domain asks (law, research, domain RAG, etc.): all AIs consult [DOMAIN_VAULT] before inventing.
Ops / Karpathy notes: [OPS_VAULT] (raw/ → wiki/ → indexes).

What’s inside (Camada 1)

File	Role
MANIFEST.md	Orientation + pointers (open this first)
agent_rules.md	Universal hard rules for any assistant
working-style.md	Collaboration, research, engineering, S/M/C
appsec-rules.md	Copyable fail-closed AppSec playbook
senior-implementer-instructions.md	DoD, TDD, staged writes
identity.md / brand-voice.md / stack.md / about-me.md	Skeletons with placeholders
ux-ui-*.md	Criteria + landing + mobile + intake
templates-agent/	Repo boot (CLAUDE.md, map, decisions)
EXAMPLE_ACTIVE_PROJECTS.md	Fictional board only
PLACEHOLDERS.md	Token legend

Quick start

git clone https://github.com/[YOUR_GITHUB]/agent-context-pack.git
cd agent-context-pack

1. Pick a language under docs/<lang>/
2. Replace placeholders from PLACEHOLDERS.md in a private fork or overlay
3. Point agents at MANIFEST.md + agent_rules.md
4. Wire [OPS_VAULT] and [DOMAIN_VAULT] to your real vault paths (private)
5. Never publish the filled private copy

Privacy & security posture

• No credential folders in this tree
• No private vault contents
• No real client/project board (example only)
• Secrets represented as [REDACTED] in rules
• Authorization is server-side; UI checks do not count
• Leak-scan before every public release (latest local scan: 0 identity hits)

Repo layout

.
├── README.md              ← you are here (hub + badges)
├── PLACEHOLDERS.md
├── LICENSE                ← MIT
├── assets/banner.svg      ← header / social art
├── templates-agent/       ← shared boot templates
└── docs/
    ├── pt/ …              ← full Camada 1
    ├── en/ …
    ├── es/ …
    └── zh/ …

Suggested GitHub topics (after publish)

agents · llm · cursor · claude · prompt-engineering · appsec · obsidian · knowledge-management · fail-closed

Roadmap for the public page (optional)

• Live shields: stars / last-commit / repo-size
• Social preview PNG from assets/banner.svg
• Discussions enabled for Q&A
• Short “diff your private overlay” guide

License

MIT — see LICENSE. Use freely; keep your private data private.