Autonomous continuous alignment and self-evolving project intelligence engine. Distills session transcripts, enforces a 200-line budget on AGENTS.md, manages path-scoped rules, records living ADRs, and compiles roadmap visualizers via Antigravity hooks. Trigger via /align or /evolve.
Scanned 9/3/2026
Install to Claude Code
npx -y skills add ksprashu/agent-skill-forge --skill continuous-alignment --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Continuous Alignment?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/ksprashu-continuous-alignment)More formats (shields.io, HTML) on the badges page.
---
name: continuous-alignment
description: Autonomous continuous alignment and self-evolving project intelligence engine. Distills session transcripts, enforces a 200-line budget on AGENTS.md, manages path-scoped rules, records living ADRs, and compiles roadmap visualizers via Antigravity hooks. Trigger via /align or /evolve.
---
# Continuous Alignment & Project Evolution Engine
Maintain a growing, self-evolving vision, mission, direction, rules, and operational instructions for your software project as development progresses.
---
## 🎯 Goal
Prevent architectural drift and context rot by continuously distilling session insights, enforcing a strict 200-line token budget on root `AGENTS.md`, cascading path-scoped rules to `.agents/rules/*.md`, compiling living Architectural Decision Records (MADRs), and maintaining an interactive SVG roadmap.
---
## 🏗️ 4-Tier Memory Topology
```
┌────────────────────────────────────────────────────────────────────────┐
│ Tier 1: Hot Invariants (`AGENTS.md` / `GEMINI.md`) │
│ • Capped at max 200 lines (~1,500 tokens) │
│ • Fast-path commands, negative constraints, root gotchas │
├────────────────────────────────────────────────────────────────────────┤
│ Tier 2: Path-Scoped Rules (`.agents/rules/*.md`) │
│ • Subsystem rules progressively loaded via path proximity │
├────────────────────────────────────────────────────────────────────────┤
│ Tier 3: Strategic Direction (`docs/VISION.md`, `docs/ROADMAP.md`) │
│ • Living roadmap, milestone tracking, glowing SVG branching visualizer │
├────────────────────────────────────────────────────────────────────────┤
│ Tier 4: Architectural Decision Records (`.gemini/knowledge/ADRs/`) │
│ • MADR formatted records capturing context, decisions, and trade-offs │
└────────────────────────────────────────────────────────────────────────┘
```
---
## 📋 Operational Workflows
### 1. Autonomous Turn Distillation (Antigravity `Stop` Hook)
When an agent turn completes, Antigravity executes `skills/continuous-alignment/scripts/distill_session.py`:
1. Reads the turn execution `transcript.jsonl`.
2. Extracts verified user constraints, error troubleshooting fixes, and CLI commands.
3. Appends deduplicated entries to `.gemini/knowledge/memories.jsonl`.
4. Triggers rule consolidation and living roadmap synchronization.
### 2. Manual Alignment Trigger (`/align` or `/evolve`)
To manually reconcile project instructions, review living ADRs, or prune outdated rules:
```bash
# 1. Distill recent session learnings
python skills/continuous-alignment/scripts/distill_session.py
# 2. Synchronize AGENTS.md and enforce 200-line budget limit
python skills/continuous-alignment/scripts/sync_agents_rules.py
# 3. Compile living ADRs, VISION.md, and ROADMAP.md
python skills/continuous-alignment/scripts/compile_roadmap.py
```
### 3. Context Pulse Inspection (`PreInvocation` Hook)
Inspect the active workspace invariants and recommended commands injected before model turns:
```bash
python skills/continuous-alignment/scripts/sync_agents_rules.py --pulse
```
---
## 🛡️ Key Guarantees
- **Zero Third-Party Dependencies**: Pure Python standard library (`json`, `re`, `hashlib`, `tempfile`).
- **Sub-Second Execution**: Hook execution completes in $< 150\text{ms}$.
- **Zero Token Waste**: Strict 200-line budget cap prevents context bloat in root instruction files.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!