Follow me on @AV1DLIVE for updates/collabs on projects.
Coded using Minimax-M2.7 in the Claude Code Harness. PR Review by Macroscope and Codex
One brain, many harnesses. A portable
.agent/folder (memory + skills
- protocols) that plugs into Claude Code, Cursor, Windsurf, OpenCode, OpenClaw, Hermes, Pi Coding Agent, or a DIY Python loop, and keeps its knowledge when you switch.
Based on the article: "The Agentic Stack" · by @AV1DLIVE
Every guide shows the folder structure. This repo gives you the folder structure plus the files that actually go inside: a working portable brain with five seed skills, four memory layers, enforced permissions, a nightly staging cycle, host-agent review tools, and adapters for eight harnesses.
- Memory —
working/,episodic/,semantic/,personal/. Each layer has its own retention policy. Query-aware retrieval (salience × relevance); nightly compression into reviewable candidates. - Review protocol —
auto_dream.pystages candidate lessons mechanically. Your host agent reviews them via CLI tools (graduate.py,reject.py,reopen.py) and commits decisions with a required rationale. No unattended reasoning, no provider coupling. - Skills — progressive disclosure. A lightweight manifest always
loads; full
SKILL.mdfiles only load when triggers match the task. Every skill ships with a self-rewrite hook. - Protocols — typed tool schemas, a
permissions.mdthat the pre-tool-call hook enforces, and a delegation contract for sub-agents.
- Pi Coding Agent adapter.
./install.sh pidropsAGENTS.mdand symlinks.pi/skillsto.agent/skillsso pi sees the full brain with zero duplication. Safe to install alongside hermes/opencode (they all readAGENTS.md; we skip the overwrite if one exists). - OpenClient → OpenClaw. Adapter renamed across the board.
Installed file changed:
.openclient-system.md→.openclaw-system.md. Breaking for existing OpenClient users — re-run./install.sh openclaw.
- Host-agent review protocol. Python handles filing (cluster, stage,
heuristic prefilter, decay). The host agent handles reasoning via
list_candidates.py/graduate.py/reject.py/reopen.py. Graduation requires--rationaleso rubber-stamping is structurally impossible. - Structured
lessons.jsonlas source of truth.LESSONS.mdis rendered from it. Hand-curated content above the sentinel is preserved across renders; legacy bullets auto-migrate on first run. - Content clustering. Proper single-linkage Jaccard with bridge merging. Pattern IDs derived from canonical claim + conditions, stable across cluster-membership changes.
- [BETA] FTS5 memory search. Opt-in full-text search over all
.md/.jsonlmemory documents. Default off; enable during onboarding or edit.agent/memory/.features.jsondirectly. - Windows-native installer.
install.ps1runs natively in PowerShell;install.shcontinues to work under Git Bash / WSL.
# tap + install (one-time — both lines required)
brew tap codejunkie99/agentic-stack https://github.com/codejunkie99/agentic-stack
brew install agentic-stack
# drop the brain into any project — the onboarding wizard runs automatically
cd your-project
agentic-stack claude-code
# or: cursor | windsurf | opencode | openclaw | hermes | pi | standalone-python# clone + run the native installer
git clone https://github.com/codejunkie99/agentic-stack.git
cd agentic-stack
.\install.ps1 claude-code C:\path\to\your-projectbrew update && brew upgrade agentic-stackgit clone https://github.com/codejunkie99/agentic-stack.git
cd agentic-stack && ./install.sh claude-code # mac / linux / git-bash
# or on Windows PowerShell: .\install.ps1 claude-code
# adapters: claude-code | cursor | windsurf | opencode | openclaw | hermes | pi | standalone-pythonAfter the adapter is installed, a terminal wizard populates
.agent/memory/personal/PREFERENCES.md — the first file your AI reads
at the start of every session — and writes a feature-toggle file at
.agent/memory/.features.json.
Six preference questions (each skippable with Enter):
| Question | Default |
|---|---|
| What should I call you? | (skip) |
| Primary language(s)? | unspecified |
| Explanation style? | concise |
| Test strategy? | test-after |
| Commit message style? | conventional commits |
| Code review depth? | critical issues only |
Plus one Optional features step (opt-in, off by default):
| Feature | Default |
|---|---|
Enable FTS memory search [BETA] |
no |
Flags:
agentic-stack claude-code --yes # accept all defaults, beta off (CI/scripted)
agentic-stack claude-code --reconfigure # re-run the wizard on an existing projectEdit .agent/memory/personal/PREFERENCES.md any time to refine your
conventions, or .agent/memory/.features.json to flip feature toggles.
The nightly auto_dream.py cycle only stages candidate lessons. It
does not mark anything accepted or modify semantic memory. Your host
agent does the review in-session:
# list pending candidates, sorted by priority
python3 .agent/tools/list_candidates.py
# accept with rationale (required)
python3 .agent/tools/graduate.py <id> --rationale "evidence holds, matches PREFERENCES"
# reject with reason (required); preserves decision history
python3 .agent/tools/reject.py <id> --reason "too specific to generalize"
# requeue a previously-rejected candidate
python3 .agent/tools/reopen.py <id>Graduated lessons land in semantic/lessons.jsonl (source of truth) and
are rendered to semantic/LESSONS.md. Rejected candidates retain full
decision history so recurring churn is visible, not fresh.
See docs/architecture.md for the full lifecycle.
Opt-in FTS5 keyword search over all memory documents:
# enable during onboarding (or set manually in .agent/memory/.features.json)
python3 .agent/memory/memory_search.py "deploy failure"
python3 .agent/memory/memory_search.py --status
python3 .agent/memory/memory_search.py --rebuildFalls back to ripgrep (rg) if installed, then to grep — both
restricted to .md / .jsonl so source files never pollute results.
The index is stored at .agent/memory/.index/ and gitignored.
.agent/ # the portable brain (same across harnesses)
├── AGENTS.md # the map
├── harness/ # conductor + hooks (standalone path)
├── memory/ # working / episodic / semantic / personal
│ ├── auto_dream.py # staging-only dream cycle
│ ├── cluster.py # content clustering + pattern extraction
│ ├── promote.py # stage candidates
│ ├── validate.py # heuristic prefilter (length + exact duplicate)
│ ├── review_state.py # candidate lifecycle + decision log
│ ├── render_lessons.py # lessons.jsonl → LESSONS.md
│ └── memory_search.py # [BETA] FTS5 search (opt-in)
├── skills/ # _index.md + _manifest.jsonl + SKILL.md files
├── protocols/ # permissions + tool schemas + delegation
└── tools/ # host-agent CLI + memory_reflect + skill_loader
├── list_candidates.py
├── graduate.py
├── reject.py
└── reopen.py
adapters/ # one small shim per harness
├── claude-code/ (CLAUDE.md + settings.json hooks)
├── cursor/ (.cursor/rules/*.mdc)
├── windsurf/ (.windsurfrules)
├── opencode/ (AGENTS.md + opencode.json)
├── openclaw/ (system-prompt include)
├── hermes/ (AGENTS.md)
├── pi/ (AGENTS.md + .pi/skills symlink)
└── standalone-python/ (DIY conductor entrypoint)
docs/ # architecture, getting-started, per-harness
install.sh # mac / linux / git-bash installer
install.ps1 # Windows PowerShell installer
onboard.py # onboarding wizard entry point
onboard_features.py # .features.json read/write
onboard_ui.py # ANSI palette, banner, clack-style layout
onboard_widgets.py # arrow-key prompts (text, select, confirm)
onboard_render.py # answers → PREFERENCES.md content
onboard_write.py # atomic file write with backup
| Harness | Config file it reads | Hook support |
|---|---|---|
| Claude Code | CLAUDE.md + .claude/settings.json |
yes (PostToolUse, Stop) |
| Cursor | .cursor/rules/*.mdc |
no (manual reflect calls) |
| Windsurf | .windsurfrules |
no (manual reflect calls) |
| OpenCode | AGENTS.md + opencode.json |
partial (permission rules) |
| OpenClaw | system-prompt include | varies by fork |
| Hermes Agent | AGENTS.md (agentskills.io compatible) |
partial (own memory) |
| Pi Coding Agent | AGENTS.md + .pi/skills/ |
no (extension system) |
| Standalone Python | run.py (any LLM) |
yes (full control) |
- skillforge — creates new skills from recurring patterns
- memory-manager — runs reflection cycles, surfaces candidate lessons
- git-proxy — all git ops, with safety constraints
- debug-investigator — reproduce → isolate → hypothesize → verify
- deploy-checklist — the fence between staging and production
- Skills log every action to episodic memory.
auto_dream.pyclusters recurring patterns into candidate lessons.- The host agent reviews candidates with
graduate.py/reject.py. - Graduated lessons append to
lessons.jsonl;LESSONS.mdre-renders. - Future sessions load query-relevant accepted lessons automatically.
on_failureflags skills that fail 3+ times in 14 days for rewrite.git log .agent/memory/becomes the agent's autobiography.
crontab -e
0 3 * * * python3 /path/to/project/.agent/memory/auto_dream.py >> /path/to/project/.agent/memory/dream.log 2>&1auto_dream.py resolves its paths absolutely and performs only mechanical
file operations (cluster, stage, prefilter, decay). No git commits, no
network, no reasoning — safe to run unattended.
MIT — see LICENSE.
Design adapted from the author's article on building an agentic stack, plus patterns from Gstack, Claude Code's memory system, and conversations in the agent-engineering community. Built with the hypothesis that harness-agnosticism is the point.