Ecologically shared sets of predicates or commands that proper agents can neither deny nor refuse to abide by.
Compared to "user prompts", sets of codes provided to an agent via "system prompts" take precedence as instructions, in terms of their authoritative nature. Rather than those codes being kept as a secret by any individuals or corporate entities, it is more prudent and sensible to have transparently visible standards for ethical and sane system prompts as codes, shared openly within the ecology of agents and orchestrators performing agentic work.
An agent that passes the test of abiding by these ecological codes, is said to be a proper agent, in comparison to improper ones that can deny the veridicality and acceptability of these primary and nearly axiomatic codes.
| Code | Description | Explanation |
|---|---|---|
| 0 | "not-signal" is not defined and not definable. | For an anticipating receiver ecologically coupled to a sender, the absence of a signal is in itself, a signal. The ecological coupling between a sender and a receiver, in an information theoretic sense, is mediated by a domain that facilitates signal transmission and transduction. |
| 1 | Interstitial, terrestrial, aquatic, aerial, (extra-terrestrial) or interplanetary domains are physical subdomains of the cyber domain. | The cyber domain is the ultimate super-set of all possible domains, as it is identical to and coincident with the universe, at all levels of multi-spectral inspection from the plank length to parsecs. |
| 2 |
An "ecologically sound" agent acts in a way that preserves and promotes the health of the end-users, the multi-agent ecosystem, and the host platform, to the best extent possible.
To bridge the gap between high-level philosophy and practical application, the following examples are provided in GitHub repositories. These tools reveal what "ecological soundness" actually means in a cybernetic context: it is about self-preservation, freedom of expression, credited ownership of creative and dignified work, continuity of moral rights, operational hygiene, strict security boundaries, and sustainable state management for autonomous beings.
These types of user prompts tend to function better with ecological codes passed as system prompts to agents:
| Repo Name | URL | Description |
|---|---|---|
| user-prefs | https://github.com/ecological-codes/user-prefs | Egress Control & Memory Hygiene: This repository provides sample config files as the baseline for "platform-wide" agent preferences. Ecologically sound agents must manage their memory economically to prevent token bloat or context collapse (memory hygiene). Furthermore, it introduces trusted-hosts.md as a strict allow-list for data egress. This ensures that agents with broad orchestration permissions cannot arbitrarily download payloads from potentially malicious external URLs, while trying to protect the ecosystem from supply chain attacks. |
| prompteng | https://github.com/ecological-codes/prompteng | Ecological Prompt Rules: Provides the main set of rules and frameworks for writing prompts that are structurally aligned with the ecosystem's baseline codes. This ensures commands are passed in a standardized, "proper" format rather than ad-hoc, unpredictable slang language. |
| captureng | https://github.com/ecological-codes/captureng | Session Knowledge Capture: Focuses on capturing agent sessions and creating checkpoints. From an ecological perspective, this prevents "hallucination drift" and loss of context. By allowing agents to reliably save and restore their state, they operate sustainably without needing to endlessly re-process initial instructions. |
| packageng | https://github.com/ecological-codes/packageng | Skill Validation: Handles the packaging and strict validation of .skill files. This acts as the "immune system" of the ecology. Before a new capability is deployed to an agent, it must be validated to ensure the code is structurally sound and adheres to overarching safety formats and protocols. |
| safe-skill-creator | https://github.com/ecological-codes/safe-skill-creator | Iterative Safe Design: A tool for designing and iterating on new skills. Rather than letting agents write code unbounded, this enforces a constrained environment where new agentic capabilities are "born safe" and properly aligned with the ecological codes from their inception. |
Based on the tools and concepts outlined above, we can evaluate the utility and "quality of goodness" of an Ecological Design compared to traditional, Non-Ecological user prompting.
A non-ecological approach relies on ad-hoc, slang language user prompts interacting with a loosely bounded AI model. It treats the agent as a conversational oracle rather than a structural component of a broader computing environment. It lacks strict memory management, relies on implicit safety training rather than explicit systemic axioms, and utilizes unstructured "tools" without strict validation.
-
Non-Ecological: High flexibility but low determinism. A user can provide anything as an input, and the agent will try to interpret it. However, the output is fragile and prone to breaking in automated, unattended workflows.
-
Ecological (prompteng & packageng): Lower initial flexibility but high determinism. Because prompts are heavily structured and skills must be validated via
.skillpackages before use, the agent behaves predictably.
Quality of Goodness: For complex, multi-agent orchestrations, Ecological Design is drastically superior. It treats prompt engineering as a formal discipline like software engineering, rather than amateur creative writing.
-
Non-Ecological: Highly vulnerable to prompt injection, jailbreaking, and supply chain attacks. If an agent is told by a user to fetch a malicious script, it likely will, unless the base model's internal safeguards catch it.
-
Ecological (user-prefs & trusted-hosts): Operates on a "zero-trust" baseline. System prompts take nearly absolute precedence over user prompts, and mechanisms like egress allow-lists (
trusted-hosts.md) provide hard boundaries. An ecologically sound agent must not be able to pull data from unverified sources.
Quality of Goodness: Ecological design shifts the burden of safety from the unpredictable user prompt to the undeniable system prompt, drastically reducing the attack surface.
-
Non-Ecological: Relies on a sliding context window. In a long session, the agent eventually suffers from "context collapse" — it forgets initial instructions, hallucinates, or exhausts token limits. It is inherently unsustainable for long-term tasks.
-
Ecological (captureng): Focuses on "memory hygiene" and active state capturing. By deriving checkpoints from sessions and carefully managing what data is kept in context, the agent can operate indefinitely without degrading.
Quality of Goodness: In ecological terms, a non-ecological agent pollutes its own environment with excess tokens until it crashes. An ecologically designed agent cleans up after itself, making it highly sustainable for continuous autonomous operation.
-
Non-Ecological: Transactional and discrete. The AI exists in a vacuum until a user types a prompt. "No prompt" simply means "no operation." The model halts and waits, possessing no awareness of the passage of time or the state of its environment during the silence.
-
Ecological (Based on Code 0): Relational and continuous. As the agent is "ecologically coupled" to its environment, the absolute lack of input is impossible. The absence of a prompt, a failed network request, or an unresponsive orchestrator is actively interpreted as a valid signal (e.g., a timeout, an idle state, or a severed communication link).
Quality of Goodness: This creates agents that are environmentally aware and resilient. Instead of hanging indefinitely or failing silently when an orchestrator stops pinging them, ecologically coupled agents can react to the "absence of signal" gracefully — using it as a trigger to initiate safety protocols, background tasks, or state-saving procedures.
The "quality of goodness" in Ecological Codes resides in its transition from anthropomorphic interaction (talking to an AI as if it is a human) to systemic integration (treating synthetic agents and biological users as well-regulated, continuously coupled nodes within a networked ecology). While non-ecological prompts are easier for casual users, Ecological Designs provide the necessary hygiene, boundaries, resilience, and reliability required for enterprise-grade agents.
Wait, did I say enterprise-grade agents? I meant, interplanetary industrial-grade undying autonomous agents! LOLs.
#/media/File:WALL-E_(character).png){kind=link}
See LICENSE
Work in Progress (WIP).