
vigil-soc
Work your team's Vigil from Claude Code: investigations and hunts that run in your own SOC, shared with colleagues.
- GitHub stars
- 443
Install
/plugin install vigil-soc@vigilRun it inside Claude Code after adding the marketplace with /plugin marketplace add Vigil-SOC/vigil.
Contents
Discovered from sourceSkills 7
Case
Give the analyst one case, $ARGUMENTS, and its evidence as a brief.
Vigil's kernel does the work. This session starts it, steers it and reads it: it does not reason through a workflow itself, and it never acts on Vigil except through the vigil MCP tools.
Read it with get_case, get_case_iocs and get_case_comments; get_case_tasks and list_runs with the case id when tasks or past runs bear on it. Lead with what the case is about, its status and severity, then the findings, IOCs and the latest comments. Cite ids as Vigil returns them; do not invent evidence the tools did not return. If no case id was given, ask for one or offer /vigil-soc:queue.
Changing the case (comments, IOCs, status, closing) goes through the case write tools only when the analyst asks for that change; each is recorded on the trail under the credential's person.
Hunt
Have Vigil hunt for the hypothesis in $ARGUMENTS and bring back what it found.
Vigil's kernel does the work. This session starts it, steers it and reads it: it does not reason through a workflow itself, and it never acts on Vigil except through the vigil MCP tools.
Choose the hunt workflow from list_workflows (or an operator agent from list_operator_agents), and ask when more than one fits. A run starts on a case: use the case the analyst names, or ask whether to open one with create_case. Start it with start_run, the case id and the hypothesis, and give the analyst the run id.
Read progress with get_run, steer with send_directive only when asked, and point to /vigil-soc:waiting if it pauses on an approval. Once it has ended, report its conclusion, citations and declared gaps as Vigil gave them. list_completed_hunts lists earlier hunts when the analyst asks what was already hunted.
Investigate
Have Vigil investigate the case in $ARGUMENTS and bring back what it concluded.
Vigil's kernel does the work. This session starts it, steers it and reads it: it does not reason through a workflow itself, and it never acts on Vigil except through the vigil MCP tools.
Choose what to start from list_workflows (an investigation workflow) or list_operator_agents (an operator agent starts as agent-run-<id>); ask when more than one fits. Start it with start_run and the case id, passing the analyst's question as context or hypothesis. Give the analyst the run id.
While it runs, get_run shows its status. Steer it only when the analyst asks, with send_directive (a note, a lead, extend, conclude or abort). If it pauses on an approval, say what it waits on and point to /vigil-soc:waiting.
Read the conclusion only after the run has ended: get_run then returns the conclusion, disposition, citations and declared gaps. Report them as Vigil gave them, gaps included; do not fill a gap with your own reasoning. A run that ends failed or cancelled is reported with its reason.
Queue
Show the analyst the open findings and the open cases in their Vigil, newest and most severe first.
Vigil's kernel does the work. This session starts it, steers it and reads it: it does not reason through a workflow itself, and it never acts on Vigil except through the vigil MCP tools.
Use list_findings and list_cases. Narrow by severity, status or time when the analyst asks. Keep it to a short table per kind: id, title, severity, status and age. Offer to open a case with /vigil-soc:case or start an investigation with /vigil-soc:investigate. Report a refusal or error as given.
Status
Find out whether the vigil MCP server answers for this credential.
Vigil's kernel does the work. This session starts it, steers it and reads it: it does not reason through a workflow itself, and it never acts on Vigil except through the vigil MCP tools.
Make one cheap read, such as list_findings with a limit of 1. If it answers, say that Vigil answers for this credential and what the read returned. Do not guess which person the credential belongs to.
A refusal or an error is the answer, not something to work around. Report it as given. The usual causes:
- the
vigilserver is missing from/mcp: the plugin's Vigil address or credential is unset (/plugin→ vigil-soc → configure); - 404: the operator has not opened the MCP surface;
- 401: the credential is wrong, expired or revoked; mint a new one;
- a permission refusal: the credential's person lacks that role.
Trail
Show the analyst the part of Vigil's audit trail they ask about.
Vigil's kernel does the work. This session starts it, steers it and reads it: it does not reason through a workflow itself, and it never acts on Vigil except through the vigil MCP tools.
Use list_audit_entries, filtering by person, actor, area, object, kind or time as asked. It returns entries oldest first with a tip; page on after with the last seq read to get newer ones. Summarise as a short timeline: time, who (person and actor), what, on which object, outcome.
Reading the trail needs audit.read, which Admin, Manager and auditor roles hold. An analyst's credential is refused; report that refusal as the answer and say a person with one of those roles can read it.
Waiting
Show the analyst what waits on a person: pending approval actions and the runs paused on them.
Vigil's kernel does the work. This session starts it, steers it and reads it: it does not reason through a workflow itself, and it never acts on Vigil except through the vigil MCP tools.
get_run on a paused run lists the approval ids it waits on; list_approval_actions lists pending actions across Vigil, and get_approval_action reads one in full. Say for each what it would do, to what, who or what proposed it, and why.
This client cannot approve. approve_action over MCP leaves the action pending, by design: approving happens in Vigil's Approvals screen, by a signed-in person. Say so plainly rather than retrying. Rejecting is allowed: use reject_action with the analyst's reason, only when they ask.
MCP servers 1
Sponsors
Details
- Marketplace
- Vigil-SOC/vigil
- Source
- clients/claude-code
- Category
- Community
- License
- Apache-2.0