Fewer reworks
Align with the real project baseline, owners, and constraints before editing instead of expanding from guesses.
Open-source Governance Engineering Method Pack
Give your Agentengineering judgmentbefore action.
Aegis brings baseline-first work, change necessity, systematic diagnosis, anti-entropy, and evidence-backed delivery into the Coding Agent you already use.
Reduce rework, rely less on blind “done” claims, and keep simple tasks simple.
Why Aegis
Put engineering judgment into how the Agent works.
A prompt can state a goal, but it rarely maintains project facts, change boundaries, and completion evidence over time. Aegis organizes those needs into engineering methods that enter only when relevant.
Safer changes
Test whether a change is needed, bound its impact, and track old paths that must retire.
Proof before done
Support completion with fresh checks, covered scope, and residual risk.
Simple stays simple
Keep trivial work on a fast path and expand governance only when risk or complexity warrants it.
Method Pack
Composable engineering methods, selected by the work.
Aegis routes to relevant methods instead of forcing every task through one fixed ceremony.
Baseline first
Read goals, constraints, owners, and acceptance before deciding how to act.
Change necessity
Confirm that a modification solves the observed problem instead of changing code for its own sake.
Anti-entropy governance
Reduce duplicate owners, retired fallbacks, and uncontrolled expansion.
Systematic diagnosis
Use observable facts to find causes before selecting the smallest effective repair.
Evidence-backed delivery
Separate actions that ran, facts established by verification, and remaining unknowns.
Long-task continuity
Keep multi-turn work aligned through durable goals, plans, and context.
Across Hosts
Methods follow the project instead of one Agent.
Aegis targets Coding Agent hosts that support Skills or installable method packs. Codex and OpenCode have fresh evidence for the current method-pack scope; use the compatibility matrix for the current status of other hosts.
View the host compatibility matrix- Codex
- OpenCode
- Claude Code
- DeepSeek Harness
- Kimi Code CLI
- GitHub Copilot
Public Evidence
Open methods, tests, and bounded benchmark evidence.
The Aegis repository publishes the method pack, install verification, host guides, and a frozen A/B benchmark. The benchmark reports observations for stated versions and conditions; it is not a universal correctness or completion-authority claim.
Aegis × Autoloom
A method pack and a product runtime solve different layers.
Bring methods to an existing host
For developers who want Governance Engineering in tools such as Codex or OpenCode. The Aegis repository owns installation, updates, compatibility status, and complete documentation.
GitHubWeave governance into execution
For users who need governance checkpoints, full trajectory, structured records, and a Windows desktop workflow. Autoloom includes the integration; no separate Aegis installation is needed.
Explore AutoloomKeep the capability boundaries clear.
- Not an Agent runtime
- Aegis does not execute model requests, file operations, or background jobs.
- Does not grant permission
- The active host and user configuration still own permissions, sandboxing, and tool access.
- Does not guarantee correctness
- Methods and evidence support review; they do not replace tests, user acceptance, or professional judgment.
- Project rules take precedence
- User instructions and target-project rules outrank general Aegis guidance.