Netra Skills — advisory security workflows you can inspect before you trust them

Netra Skills are repeatable, governed workflows that run on VulnTrek's own read-only tools inside a single tenant. Each skill states its scope, its execution tier and the evidence it used, so a reviewer can check the reasoning instead of accepting a verdict.

Last reviewed:

Triage daily queue (Supervised (approve each action))

Scope: One capped batch of the organization's open findings, scored across CVSS, EPSS, CISA KEV, asset criticality and exposure, with duplicates grouped and exception-covered findings flagged. Example questions: Which of last night's findings belong in P0, and on what evidence? Show the factor breakdown behind this priority band. Group the duplicates in this scan drop before anyone triages them by hand.

Tenant advisory agent (Advisory (read-only))

Scope: Only this organization's own history: the open finding register, deduplication merges and reversals, remediation speed, due-date reliability and how humans previously accepted, modified or rejected advice. Example questions: Where is our remediation consistently slipping, and what should we change first? Which advice did our team reject last quarter, and did the outcome justify it? What patterns in our own record predict a missed due date?

Finding explanation agent (Advisory (read-only))

Scope: One authorized finding in the caller's active organization, explained from directly linked evidence only. Closed-world: if the evidence is not linked, the skill says so rather than inferring. Example questions: Why does this finding exist, and why is it prioritised this way? Which evidence drove this score, and which factor dominated? What would change this priority if it were no longer true?

Remediation SLA check (Advisory (read-only))

Scope: Remediation due dates for this organization: past due, CISA KEV past the KEV due date, due soon, past due with no owner, and open findings with no due date at all. Example questions: What breaches SLA today, and who owns it? Which KEV findings are past their federal due date? Which past-due findings have no owner assigned?

Release gate (Supervised (approve each action))

Scope: One build or change, evaluated against the existing CI gate policy plus tenant thresholds, returning GO, WARNING or BLOCKED with the deciding rule and the triggering findings. Example questions: Can this release ship, and which rule decided that? Which findings are blocking the gate right now? Draft the escalation for a blocked release so a human can send it.

Executive risk brief (Advisory (read-only))

Scope: This organization's posture score and P0–P3 mix, opened versus closed velocity, SLA-met share, top exposure movers and KEV exposure, with rule-based recommended actions. All figures are modelled estimates. Example questions: Write this week's board-ready risk brief from our own numbers. What moved most in our exposure ledger since last week? Give me three defensible actions for the leadership meeting.

Audit evidence pack (Advisory, no learning)

Scope: A control-by-control evidence index over a rolling window, covering ISO/IEC 27001:2022 Annex A A.8.7 and A.8.8 with deterministic checks and Covered / Partial / Gap status. An evidence index, never a compliance attestation. Example questions: Assemble the evidence an auditor will ask for on technical vulnerability management. Which controls are only partially evidenced in our own records? Export the supporting findings behind this control as a pack.

Incident evidence report (Advisory, no learning)

Scope: The full Pramana timeline for one finding — lifecycle dates, audit events and other findings on the same asset — drafted into a CERT-In-style submission and an internal technical report. Drafts are never submitted. Example questions: Reconstruct what we knew about this finding, and when. Draft the regulator submission for review against the six-hour clock. What else on this asset belongs in the incident narrative?

Posture snapshot (Advisory (read-only))

Scope: The organization posture summary and current remediation campaigns pulled through the read-only MCP tools — the foundation skill used to verify the runner, kill switch and dry-run mode end to end. Example questions: Give me the current posture summary for this organization. Confirm the skill runner, kill switch and dry-run mode behave as documented. Which campaigns are open right now?

Advisory (read-only)

The skill reads tenant records and returns a recommendation. Nothing is written except the run record itself.

Supervised (approve each action)

Accepting a recommendation routes it through the existing approval path, one human decision per action.

Advisory, no learning

Used where output format must not drift — the skill takes no feedback and never adapts between runs.

Dry run before live run

Every skill can execute all of its steps without recording anything to the Pramana trail, so a reviewer can see the output before it becomes evidence.

Step log, not a black box

Each run publishes its ordered steps, the tool each step called, its duration and its result summary.

Tenant boundary enforced in the database

Skills read through org-scoped policies and security-definer functions, so a skill cannot reach another organization's records even if asked to.

Kill switch and per-organization settings

Org admins can tune or disable any skill for their organization, and the platform can disable a skill everywhere.

What Netra Skills never do

Netra Skills recommend; people decide. They read only the caller's active organization, never learn across tenants, and never close a finding, assign an owner, change a score, grant an exception or dispatch a change on their own. Every live run is written to the append-only Pramana trail, and every suggested follow-up needs approval by an authorized human.

Can a Netra Skill close a finding or apply a fix?

No. Netra Skills are advisory and read-only. Accepting a recommendation routes it through the existing approval path, where an authorized human makes the decision that is recorded.

Does Netra learn from one customer and apply it to another?

No. The tenant advisory agent learns only from that organization's own history, and no shared model is trained on customer data. There is no cross-tenant learning of any kind.

How do I check what a skill actually did?

Each run publishes an ordered step log with the tool called at every step, and every live run is written to the append-only Pramana trail with the inputs and output attached.

What is the difference between Netra Skills and Netra Actions?

Netra Skills read and recommend. Netra Actions are the governed remediation agents that may dispatch catalogued actions under an execution tier, with one human approval per action and published caps.

Is Ask Netra the same thing as a Netra Skill?

Ask Netra is the conversational door onto the same read-only tools. It answers questions about your own findings and cites the record it used; it is not a general-purpose chatbot and it never closes a finding.

Can we use these skills from our own AI client instead?

Yes. The same read-only tools are exposed over VulnTrek's MCP server, so an authorized MCP client works against the same tenant scope and the same governance rules.