The safeguards map / 3.2 Agent harnesses
Harness²
Drafted with AI
Drafted with AI from the cited sources under the direction of UnGovr staff. UnGovr holds editorial responsibility for this page. How this site is made
A research framework that proposes, probes and composes edits to an agent harness, guardrails among its eight editable components.
Where it sits
3.2 Agent harnesses (primary): Optimizes an agent harness by proposing edits to its components, including guardrails, and keeping the refined harness; it does not itself limit an agent.
Harness² exposes eight editable components around the model: **system prompt, guardrails, memory, tool guidance, subagents, skills, scripts, and plugins**.
captured
propose edits, probe their effects on the same task, and compose a refined harness from behavioral contrasts between executions
captured
Facts
- Concern
- safety
- Names the law
- no
- Subject
- agent
- Used by
- those who make it
- Runs
- by hand
- Policy comes from
- owner-entered settings
- Can the agent bypass it
- depends
Handoffs
None recorded.
Listed, not drawn: it is not yet released, and no documented handoff and no foundation home places it on the map.
Every quote is verbatim from a stored capture of the page linked under it. Placement is our reading of the product's own documentation; no publisher has reviewed or endorsed it. Rows CC BY 4.0. Names, marks and logos belong to their owners; logos identify each product and are not part of the dataset.