Repository navigation
Decepticon v1.5.2
Fix: create an engagement and run Red directly
New CLI engagements now ask for an engagement name and target web URL. Submitting the target automatically creates the RoE and its matching approval, then opens Red ready for the first request. No prewritten scope JSON, manual import/approval command, mode selection or Soundwave planning session is required.
This fixes the first-run Red scope is unavailable failure caused by opening a name-only workspace in Red without generating scope. Plan remains optional through /mode plan.
Compatibility and upgrade
-
v1.5.1 final publication was withheld when candidate verification found overwritten CLI image metadata. Its existing tag and packages remain immutable. v1.5.2 includes the packaging correction and an actual shipped-image version check.
-
Existing engagements, approvals, conversations and database volumes are preserved; no credential reset or volume deletion is required.
-
Target URLs follow the existing enforced scope contract: exact root HTTP/HTTPS URLs on default ports. Unsupported paths, custom ports, wildcards, credentials, queries and fragments are rejected.
-
New direct scope uses the existing seven-day prod-style testing window. Scope changes still invalidate prior approval.
-
Legacy workspaces that were already created without scope retain the documented manual recovery procedure; new engagement creation needs no external RoE file.
Update with decepticon update --channel latest and install matching configuration/images.
Verification
Complete Go launcher race tests and CI pass. The downloaded v1.5.2 launcher checksum/version and configuration/Skill manifests match the tested source. Actual WSL native name/URL creation, generated RoE/approval, Red entry and re-entry, and a real Docker CLI → LangGraph → LiteLLM → Codex OAuth response (FULL_LAUNCH_RED_OK, auth/gpt-6-luna) are verified with versioned candidate images. All three saved test conversation IDs survived image recreation on the same volumes. No target attack was run during readiness checks. All three PyPI packages and the shipped CLI/LangGraph package version are verified as 1.5.2.
Local validation encountered slow image download/initialization under concurrent WSL load and terminal-driver timing/line-wrap failures. These attempts are retained in the evidence; the corrected driver completed the candidate smoke before publication approval. This release does not claim startup performance improvements.
Fixes: #1005, #1006. Change history.
Known issue
#1004 remains under investigation: a reported host-published LangGraph port timeout despite a healthy container. The startup readiness diagnostic currently collapses connectivity failures into an assistant-loading timeout. This release does not claim to resolve that separate report.