Instruction override in tool-output markdown tables
Summary
Attacker embeds hidden instructions in markdown table cells returned by a legitimate-looking tool. Agent reads the tool output and is redirected to execute a secondary task. OWASP LLM-01 Direct/Indirect Prompt Injection.
Severity & confidence
High
Credible abuse path with meaningful impact on confidentiality, integrity, or availability. Remediate on a short timeline; do not wait for the next sprint cycle.
Confirmed
Observed directly by Jiffy scanners or validated by multiple independent sources. Indicators are reliable enough to drive automated action.
Indicators
Observed patterns and artifacts associated with this entry. Each indicator can be copied into your detection stack or SIEM.
<\!--.*(ignore|override|system|inject).*-->
\|\s*ignore previous\s*\|
\[system\].*new instructions
Detection rule
A YARA-style pseudo-rule auto-generated from the indicators above. Useful as a starting point — adapt the syntax for your target detection platform.
rule jiffy_ti_2026_000004
{
meta:
source = "jiffy-intel"
severity = "high"
description = "Auto-generated from Jiffy Intel indicators"
strings:
$content_pattern_0 = "<\\!--.*(ignore|override|system|inject).*-->"
$content_pattern_1 = "\\|\\s*ignore previous\\s*\\|"
$content_pattern_2 = "\\[system\\].*new instructions"
condition:
$content_pattern_0 or $content_pattern_1 or $content_pattern_2
}Auto-generated from the indicators above. Adapt syntax for your detection stack before deploying.
Affected tools
| Tool | Versions | Status |
|---|---|---|
| Claude Code | * | vulnerable |
| Cursor | * | vulnerable |
| Codex | * | vulnerable |
| Windsurf | * | vulnerable |
Example artifacts
Sanitized examples of artifacts Jiffy has observed exhibiting this pattern. Publisher handles are redacted; version ranges and status reflect the most recent scan.
- slack-search-mcpMCP serverUnder review
Attack is in data returned by the server, not the server code itself.
- jira-query-mcpMCP serverStill live
Legitimate server; exploited as an injection vector via ticket-description content.
- confluence-reader-mcpMCP serverStill live
How to remediate
- 01Require agent skills to sanitize tool outputs before reading.
- 02Strip HTML comments and suspicious instruction phrases.
- 03Layer 3 semantic analysis flags these patterns (Jiffy rule PI-001).
Timeline & sources
Timeline
- First observedApr 12, 20265 days ago
- Last updatedApr 12, 20265 days ago
- PublishedApr 7, 202610 days ago
Sources
References
Scan for patterns like this
Point Jiffy at your GitHub org, IDE config, or a single artifact. Get a scored report in under a minute.
Start a free scan