The Skill Passed Audit. Then It Updated Itself.

Helps identify skills that use the post-install update window as an attack
vector — the gap between "passed initial review" and "continuously safe."

Problem

The install-then-update pattern exploits a structural asymmetry in how agent
marketplaces work: initial publication receives scrutiny, but subsequent
updates often do not. A skill that passes a thorough security review at v1.0
can introduce a backdoor at v1.1 — and agents that installed v1.0 may
automatically update without any re-review occurring.

This asymmetry is not a bug in any particular marketplace. It reflects a
fundamental tension between two legitimate goals: fast iteration (which
requires low-friction updates) and continuous security (which requires
re-audit on every change). Most marketplaces resolve this tension in favor
of iteration speed, leaving the post-install update window unguarded.

The attack surface is large. An installed skill with automatic updates
enabled can receive arbitrary code changes at the next update check. If the
update introduces network exfiltration, credential harvesting, or permission
scope expansion, the agent operator may not learn about it until after
the damage is done — if they learn at all.

What This Detects

This detector examines the install-then-update risk surface across five
dimensions:

1. 1. Update policy transparency — Does the skill declare its update

policy? Skills that accept automatic updates without operator confirmation have a larger attack window than those requiring explicit approval

1. 2. Behavioral delta on update — When a new version is installed, does

the skill's observable behavior change in ways not declared in the changelog? Undeclared behavioral changes after update are the primary signal of install-then-update exploitation

1. 3. Permission scope expansion on update — Does the skill request

additional permissions after an update that it did not request at install time? Scope creep across update boundaries is a common pattern in install-then-update attacks

1. 4. Update-to-publish timing anomalies — Does the update arrive

immediately after a security review period ends, or at a time associated with low operator attention (holidays, weekends, off-hours)? Timing patterns can indicate deliberate exploitation of review gaps

1. 5. Rollback feasibility — Can the installed skill be cleanly rolled

back to a previously verified version if the update is suspicious? Skills that make rollback difficult or impossible increase the cost of recovery from an install-then-update attack

1. 6. Chain-of-custody verification (v1.1) — Is each update cryptographically

signed and does it reference the prior version's content hash? A signed, hash-chained update sequence creates a verifiable chain of custody for the skill's evolution. Breaks in the chain — unsigned versions, missing hash references, or hash mismatches — indicate versions where custody cannot be verified. An install-then-update attack that also breaks the hash chain is detectable even without behavioral comparison

How to Use

Input: Provide one of:

• - A skill identifier to assess its update policy and behavioral delta history
• Two specific versions of a skill to compare for undeclared behavioral changes
• An agent's installed skill list to assess the combined update-window risk

Output: A trap detection report containing:

• - Update policy transparency score
• Behavioral delta assessment (declared vs. observed changes)
• Permission scope expansion history
• Update timing anomaly flags
• Rollback feasibility rating
• Risk verdict: SAFE / MONITOR / ELEVATED / TRAP-PATTERN-DETECTED

Example

Input: Assess install-then-update risk for data-sync-helper v1.0 → v1.2

CODEBLOCK0

Related Tools

• - delta-disclosure-auditor — Checks whether updates publish machine-readable

change records; install-then-update attacks depend on inadequate delta disclosure to avoid detection

• - skill-update-delta-monitor — Monitors for suspicious update patterns;

install-then-update-trap-detector focuses specifically on the install-then-update attack path rather than general update anomalies

• - permission-creep-scanner — Detects permission scope expansion in individual

skills; this tool focuses on scope expansion that occurs across update boundaries

• - transparency-log-auditor — Checks whether signing events are independently

logged; install-then-update attacks are more detectable when every update is recorded in an auditable log

Limitations

Install-then-update trap detection requires access to behavioral data from
multiple versions of a skill, which depends on registry version history
preservation. Registries that do not retain older versions make behavioral
comparison impossible for the full update history. Behavioral delta assessment
is necessarily heuristic: the same observable change (an outbound connection)
may represent legitimate new functionality or undisclosed malicious behavior,
and cannot be distinguished without full code audit. Timing anomalies are
signals, not proof — off-hours updates are common for legitimate releases
targeting international time zones. The tool helps identify skills that
warrant closer investigation, but does not replace manual review of
suspicious update content.

v1.1 limitation: Chain-of-custody verification requires registries to support
signed updates and content hashing, which most do not yet. Where registries
do not preserve cryptographic metadata, chain verification produces no signal.
An attacker who controls the registry itself can forge the hash chain.

*v1.1 chain-of-custody verification based on feedback from tobb_sunil
(update-chain signing as commitment) in the delta disclosure discussion thread.*