Security Policy

Reporting a vulnerability

If you discover a security vulnerability in taskflow, please report it privately rather than opening a public issue.

• Email: bshengtao@gmail.com
• Subject: [SECURITY] taskflow — <brief description>

I aim to acknowledge reports within 72 hours and ship a fix within 7 days for confirmed vulnerabilities.

Scope

taskflow runs subagent processes, manages a file-based cache with atomic locks, and resolves interpolation expressions from user-provided DSL definitions. Areas of particular interest:

Area	Concern
Path traversal	Can a malicious flow definition escape the workspace via cwd or file writes?
Interpolation injection	Can {args.X} or {steps.ID.output} be exploited to execute arbitrary shell commands? A string-form script run containing an interpolation placeholder is rejected at validation; the array form passes interpolated values as argv (no shell).
Cache poisoning	Can a crafted fingerprint or TTL bypass freshness guarantees?
Subagent capability	Author-written flows can execute shell — via script phases and bash-enabled subagents — so a flow definition is trusted input. Model calls, and any flow generated at runtime by a model (flow{def} / ctx_spawn), are untrusted by nature.
File lock atomicity	Can a concurrent process corrupt the run index or cache through race conditions?

The runtime has intentional hardening: realpath-based path containment, runId validation, atomic writes, and stale-lock stealing. But if you find a bypass, it's a vulnerability.

Supported versions

Version	Support
v0.1.3 (latest)	✅ Active
Earlier versions	❌ Unsupported — upgrade to latest

Disclosure

After the fix is released, vulnerabilities will be disclosed publicly in a GitHub Security Advisory and noted in CHANGELOG.md. Credit will be given unless you prefer to remain anonymous.