Independent deployments will determine whether Xaidr can reliably enforce in-process security and governance controls on AI-agent actions without prohibitive integration or performance costs.

state: expiredheat: lowuncertainty: highknownscott: lowagentic-security agent-harnesses agent-observabilityDelphi Security

What is this?

Xaidr is presented as an open-source, in-process runtime security and governance system for monitoring and controlling AI-agent actions, associated in the case with Delphi Security. The supplied web snippets establish the broader need for real-time agent controls, observability, policy enforcement, and enterprise integrations, while one source warns that same-process enforcement shares the agent’s execution boundary. However, they provide no independent deployment evidence, benchmarks, integration results, or clear product details sufficient to establish Xaidr’s reliability, overhead, or provenance; even the cited root-commit text is truncated.

Why it matters to Scott

Scott already holds the substantive position in Decision Authority Infrastructure and SiloOS: agent actions require deterministic runtime gates, audit evidence, and structural containment outside the untrusted agent. Xaidr currently adds only an unvalidated in-process implementation claim, while the radar already tracks closely related enforcement tests in GraphArc and OpenCode Guardians; without deployments or benchmarks, it neither challenges Scott’s execution-boundary thesis nor changes what he should build.

ip:framework.decision-authority-infrastructureip:framework.siloosip:concept.manners-vs-physicsdev:concept.deterministic-agent-control-planeradar:grapharc-runtime-agent-graph-gatesradar:opencode-guardians-tool-call-verificationradar:concept.agentic-securityradar:concept.agent-harnesses

queries asked of Scott's wikis

in-process versus isolated agent security enforcement
agent harness policy gates and tool authorization
runtime observability for coding-agent actions
tamper-resistant controls inside agent processes
agent governance integration and performance overhead
audit trails and compliance mapping for autonomous agents

Measured heat

no measured readings yet — the hourly heat pass fills this in

How the heat travelled

no chain yet — the hourly chain pass fills this in

Evidence (5) — ⭐ canonical anchor

source	object	author	score	comments
🟧 hn	Xaidr – In-process runtime security and governance for AI agents	delphiaisec	2	0
🟧 echo.github ⭐	The repository’s root commit is the earliest public artifact. Its commit message says: “Seed the open-source sensor from the open subset of	Delphi Security Inc.	—	—
🟧 hn	Show HN: Doberman: The AI watchdog that stops Claude from deleting your database	alanfuNZ	9	5
🟧 hn	Agent Control Plane: the LLM proposes, it never authorizes	Yacine_c75	1	0
🟧 hn	Show HN: Termaxa – my agent gate passed its security rig, failed a two-user test	devdoc83	1	1

Interpretation history

2026-08-20T16:40:53Z
No independent deployment, benchmark, bypass test, or integration report emerged during the observation window; adjacent projects remain repetitive context rather than evidence about Xaidr, so this specific implementation case has faded.
2026-08-18T15:43:27Z
Termaxa’s reported two-user failure sharpens multi-tenant authorization and isolation as a concrete test Xaidr must pass, but it remains an adjacent artifact with no deployment or evaluation of Xaidr itself. The core reliability, bypass-resistance, integration-cost, and overhead questions remain unanswered.
2026-08-18T15:24:03Z
evidence attached: hn.story.49346532 — Its security-gate result and two-user failure bear directly on whether agent action controls handle multi-user authorization safely.
2026-08-17T22:32:56Z
The added control-plane artifact broadens the adjacent implementation landscape but does not evaluate Xaidr itself. With no independent deployment, benchmark, bypass test, or integration-cost evidence, the case remains an unvalidated in-process enforcement claim.
2026-08-17T21:23:29Z
evidence attached: hn.story.49337491 — A first-party control-plane artifact directly bears on whether agents can be governed by separating model proposals from authorization.
2026-08-17T20:34:14Z
Doberman adds an independent runtime-watchdog artifact to the broader implementation landscape, but it is not a deployment or evaluation of Xaidr. Xaidr’s reliability, bypass resistance, integration burden, and performance overhead therefore remain untested.
2026-08-17T20:23:39Z
evidence attached: hn.story.49336757 — The released runtime watchdog is an independent artifact bearing directly on whether agent-action governance can enforce controls in practice.
2026-08-16T19:34:51Z
No independent deployment, benchmark, integration, or tamper-resistance evidence has appeared; the case remains an unvalidated implementation claim rather than a test of Scott’s runtime-governance thesis.
2026-08-16T19:31:07Z
grounded: known/low — Scott already holds the substantive position in Decision Authority Infrastructure and SiloOS: agent actions require deterministic runtime gates, audit evidence,
2026-08-16T19:28:41Z
origin walked (codex/luna, conf 0.96): anchor hn.story.49322372 -> echo.github.2fa7b0bd65 by Delphi Security Inc.
2026-08-16T19:27:29Z
case created — The first-party repository is a concrete runtime-security artifact distinct from scanners, endpoint observability, and external MCP proxies.

Decision trace

08-21 02:40expireNo independent deployment, benchmark, bypass test, or integration report emerged during the observation window; adjacent projects remain repetitive context rather than evidence about Xaidr, so this sp
08-21 02:40alert_silentThe only changes are negligible engagement updates with no substantive evidence about Xaidr; there is nothing new for Scott before the next briefing.
08-21 02:40alert_routeThe only changes are negligible engagement updates with no substantive evidence about Xaidr; there is nothing new for Scott before the next briefing.
08-19 01:43repriceTermaxa’s reported two-user failure sharpens multi-tenant authorization and isolation as a concrete test Xaidr must pass, but it remains an adjacent artifact with no deployment or evaluation of Xaidr
08-19 01:43alert_silentThe sparse first-party claim identifies a useful failure mode but provides no test details, root cause, or direct evidence about Xaidr; it can wait for routine review rather than interrupt Scott today
08-19 01:43alert_routeThe sparse first-party claim identifies a useful failure mode but provides no test details, root cause, or direct evidence about Xaidr; it can wait for routine review rather than interrupt Scott today
08-19 01:25alert_silentThe Termaxa post suggests a potentially relevant multi-user isolation failure in an agent gate, but the visible evidence provides no test details, root cause, benchmark, or deployment evidence. It rei
08-19 01:25surface_candidateThe Termaxa post suggests a potentially relevant multi-user isolation failure in an agent gate, but the visible evidence provides no test details, root cause, benchmark, or deployment evidence. It rei
08-19 01:25alert_routeThe Termaxa post suggests a potentially relevant multi-user isolation failure in an agent gate, but the visible evidence provides no test details, root cause, benchmark, or deployment evidence. It rei
08-19 01:24attachIts security-gate result and two-user failure bear directly on whether agent action controls handle multi-user authorization safely.
08-19 01:23propose_attachIts security-gate result and two-user failure bear directly on whether agent action controls handle multi-user authorization safely.
08-18 23:21sensor_dirtyengagement_update
08-18 08:32repriceThe added control-plane artifact broadens the adjacent implementation landscape but does not evaluate Xaidr itself. With no independent deployment, benchmark, bypass test, or integration-cost evidence
08-18 08:32alert_silentThe new material repeats an architecture Scott already tracks and supplies no consequential result about Xaidr’s reliability, tamper resistance, overhead, or deployment viability; it can wait for rout
08-18 08:32alert_routeThe new material repeats an architecture Scott already tracks and supplies no consequential result about Xaidr’s reliability, tamper resistance, overhead, or deployment viability; it can wait for rout
08-18 08:21sensor_dirtycomment_update
08-18 07:24alert_silentA newly linked control-plane repository appears to implement the familiar separation between LLM proposals and deterministic authorization, but the supplied evidence includes no deployment, benchmark,
08-18 07:24alert_routeA newly linked control-plane repository appears to implement the familiar separation between LLM proposals and deterministic authorization, but the supplied evidence includes no deployment, benchmark,
08-18 07:23attachA first-party control-plane artifact directly bears on whether agents can be governed by separating model proposals from authorization.
08-18 07:22propose_attachA first-party control-plane artifact directly bears on whether agents can be governed by separating model proposals from authorization.
08-18 06:34repriceDoberman adds an independent runtime-watchdog artifact to the broader implementation landscape, but it is not a deployment or evaluation of Xaidr. Xaidr’s reliability, bypass resistance, integration b
08-18 06:34alert_silentThe adjacent implementation does not validate Xaidr or alter Scott’s existing runtime-gate and structural-containment position; independent deployment results, benchmarks, or adversarial testing can w
08-18 06:34alert_routeThe adjacent implementation does not validate Xaidr or alter Scott’s existing runtime-gate and structural-containment position; independent deployment results, benchmarks, or adversarial testing can w
08-18 06:24alert_silentDoberman is another early runtime-guardrail implementation claim, but the available evidence provides no independent deployment results, benchmarks, bypass testing, integration-cost data, or tamper-re
08-18 06:24alert_routeDoberman is another early runtime-guardrail implementation claim, but the available evidence provides no independent deployment results, benchmarks, bypass testing, integration-cost data, or tamper-re
08-18 06:23attachThe released runtime watchdog is an independent artifact bearing directly on whether agent-action governance can enforce controls in practice.
08-18 06:23propose_attachThe released runtime watchdog is an independent artifact bearing directly on whether agent-action governance can enforce controls in practice.
08-17 05:34repriceNo independent deployment, benchmark, integration, or tamper-resistance evidence has appeared; the case remains an unvalidated implementation claim rather than a test of Scott’s runtime-governance the
08-17 05:34alert_silentThe only reobservation is unchanged engagement, with no substantive new evidence or consequential event; wait for an independent deployment, benchmark, or security evaluation.
08-17 05:34alert_routeThe only reobservation is unchanged engagement, with no substantive new evidence or consequential event; wait for an independent deployment, benchmark, or security evaluation.
08-17 05:31alert_silentThe public repository establishes an open-source, in-process sensor release, but provides no independent deployment, benchmark, tamper-resistance evidence, or integration-cost result that would change
08-17 05:31alert_routeThe public repository establishes an open-source, in-process sensor release, but provides no independent deployment, benchmark, tamper-resistance evidence, or integration-cost result that would change
08-17 05:31groundScott already holds the substantive position in Decision Authority Infrastructure and SiloOS: agent actions require deterministic runtime gates, audit evidence, and structural containment outside the
08-17 05:28promote_anchororigin walk conf 0.96
08-17 05:27createThe first-party repository is a concrete runtime-security artifact distinct from scanners, endpoint observability, and external MCP proxies.