StrayMark · documentation
Cognitive discipline for AI-assisted engineering
- Track every decision
- Detect drift empirically
- Audit-ready by default
Install the CLI in one line
$ curl -fsSL https://raw.githubusercontent.com/StrangeDaysTech/straymark/main/install.sh | sh
# creates repo-native governance artifacts
# ready for your first Charter…then follow our short quickstart guide →
Cognitive pairing
Knowledge humans can stand inside
StrayMark is a cognitive pairing tool: it turns project information into situated knowledge. It does not dump data for a machine to query; it builds a map a person can stand inside, see what is being decided and why, what is in motion, and where the work is going.
It also keeps humans from being outrun by the speed at which agents code and decide. For AI-augmented engineering, StrayMark keeps you in command of design and implementation decisions with verifiable context, while giving AI agents the cognitive discipline they need to stay coherent in medium and large projects.
Five workflows, one repo
Every step leaves a versioned artifact in your repo. No external systems, no implicit decisions.
From a fresh terminal to your first Charter
A 15-second recording of the real CLI. Three commands; no edits, no orchestration.
Why this exists
The industry is busy with models, guardrails, and compliance. The missing piece is upstream of all of them: the cognitive discipline of the team working with the agents. Without structure, agents drift, decisions are lost, risks go undocumented, and regulatory evidence becomes improvised. StrayMark structures the work itself — repo-native, agent-aware, audit-ready by construction.
What's in the box
Structured cognitive discipline
Charters define purpose and limits before code. AILOGs capture human/agent exchange. AIDECs record decisions and tradeoffs.
Repo-native by design
Everything lives in your git repo: artifacts, governance rules, agent directives. No external platform, no second source of truth.
Declarative agent governance
Versioned rules in STRAYMARK.md and AGENT-RULES bind agent behavior at the workflow level — not at runtime, not after the fact.
Evidence as a byproduct
EU AI Act, ISO 42001, NIST AI RMF, and GDPR mappings emerge from the same artifacts the team already produces. No parallel paper trail.
A CLI that does the work
init, validate, audit, analyze, compliance, metrics — one binary, eleven commands, deterministic outputs you can grep and pipe.
TDE: drift detection
The Transversal Debt Engine surfaces hidden coupling between charters before it compounds into incidents.
Skills for AI agents
Eleven slash-commands wrap the rituals: /straymark-charter-new, /straymark-ailog, /straymark-audit-prompt, /straymark-status. The agent drives the framework, not you.
Multi-model external audit
Three auditor CLIs (e.g. claude, copilot, gemini) read the same prompt and audit the Charter independently at the closure gate — before it ships. A calibrator deduplicates, reclassifies severity, and merges signed evidence into telemetry.
Emergent observation by design
Mandatory cross-references between documents let the agent spot stale specs and inter-charter drift on its own. Cognitive discipline raises the floor without tightening the prompt.
Ready to try it?
The quickstart guide walks you from a fresh terminal to a closed Charter with one external-audit cycle on top — six short sections, plain copy-paste-able commands, about ten minutes of reading. If you've read this far, that's where to go next.
Read the quickstart →