AgentSec Triage is alpha software. Reports about the scanner and corrections to its threat data are different workflows. Pick the section that matches the problem and do not attach secrets, live credentials, private repository content, or unredacted personal paths to a public report.

No private intake address is asserted in this repository yet. The temporary private-disclosure procedure below is intentionally conservative until the project owner configures and documents a verified channel.

A scanner vulnerability is a flaw in AgentSec itself, including target-repository code execution, a scan-root escape, unsafe symlink or reparse-point traversal, unintended network access, secret disclosure, or a denial of service that defeats the documented bounds.

Do not open a public issue when exploitation details or a proof of concept could put users at risk. Follow Private disclosure. Include the affected commit or version, platform, Python version, minimal reproduction, observed impact, and whether the reproduction contains sensitive data.

A false positive is a finding on benign evidence. A public issue may use this template after all sensitive material is removed:

Title: False positive: <detector ID>/<rule ID>
AgentSec version and database version:
Platform and Python version:
Command, including limits and --redact use:
Minimal benign fixture or public repository link:
Finding severity, confidence, path, and redacted evidence:
Why the evidence is benign:
Expected behavior:

Do not remove a confirmed IOC solely because one environment considers it benign. The underlying source and scope must be reviewed.

A false negative is documented in-scope attack evidence that an otherwise complete scan misses. If the evidence is already public and disclosure creates no new risk, a public issue may use:

Title: False negative: <campaign or technique>
AgentSec version and database version:
Platform and Python version:
Minimal safe fixture:
Primary source URL and access date:
Exact source claim the fixture represents:
Expected detector, rule, severity, and confidence:
Actual exit code, completion status, diagnostics, and findings:

If the evidence is unpublished, weaponizable, or contains a live indicator that could identify a victim, use Private disclosure instead.

Use a public issue only for public, non-sensitive corrections. Include:

Title: IOC correction: <record or campaign>
Database version and affected field:
Current value:
Proposed value or retraction:
Primary source URL, publication date, and access date:
Exact supporting or contradicting claim:
Confidence and status change requested:
Source license or redistribution note, if known:

Do not submit raw stolen data, credentials, victim identifiers, or a source that cannot legally be redistributed. Corrections require traceable evidence and a regression fixture when they affect detector behavior.

Accepted corrections and retractions preserve the earlier intelligence event. Do not rewrite or delete the original event to make the timeline look current. Add a new correction or retraction event with:

• an updated_date recording when AgentSec accepted the change;
• affected_event_ids naming every earlier event affected;
• status: corrected for a correction or status: retracted for a retraction;
• at least one reviewed source that supports the change;
• an explicit detector-coverage statement and a regression fixture when runtime behavior changes.

The intelligence builder rejects missing, unresolved, and self-referencing event targets. The generated timeline and public feed retain the relationship between the later change and every earlier event.

Until this repository documents a verified private reporting channel, do not send sensitive details to a guessed email address and do not open a public issue. Contact the project owner through an existing private channel you have already verified and ask for the current security intake method without including the vulnerability details. If no verified channel exists, retain the report and only publish a minimal request for a private contact method that reveals no exploit or victim information.

Once a private channel is established, send the affected version, impact, reproduction steps, safe proof of concept, suggested mitigation, and a contact for follow-up. Encrypt sensitive attachments using a key confirmed through that channel. No response-time promise is made for this alpha.