Long-session recovery · documentation

Recover the starting goal from a local Claude Code or Codex session.

Blind extracts a compact anchor from the first user turn, so you can inspect it and use it to restate the task before continuing a long session.

An open-source Hermes Labs project.

View source ↗Product overview Open-source catalog

§ 01

Install and verify

Pinned first workflow

Start by checking the released command is available.

$ pipx install hermes-blind==0.3.2

§ 02

Install for an agent host

Skill and extension routes

Host installation adds the skill or extension, not the hermes-blind Python command. Use hermes-blind if already available; otherwise prefer uvx hermes-blind==0.3.2 or install persistently with pipx install hermes-blind==0.3.2.

Claude Code

$ claude plugin marketplace add hermes-labs-ai/hermes-blind

$ claude plugin install hermes-blind@hermes-blind

OpenAI Codex CLI

$ codex plugin marketplace add hermes-labs-ai/hermes-blind

$ codex plugin add hermes-blind@hermes-blind

Gemini CLI

$ gemini extensions install https://github.com/hermes-labs-ai/hermes-blind --ref v0.3.2

Limit. Gemini CLI can host the skill, but Hermes Blind does not recover Gemini session logs; from Gemini, use it with a Claude Code or Codex session you name.

§ 03

Run one bounded task

Read the result

$ hermes-blind apply --latest --format auto --turn 9 --out recovery.md
sed -n '1,80p' recovery.md

Expected result. Blind prints the selected local session path to stderr and writes a recovery scaffold with the original anchor and observed session state. Review the file before sharing it.

§ 04

Reliance boundary

The CLI reads local Claude Code and Codex logs; it does not detect drift or decide when recovery is needed. Gemini CLI can host the skill but cannot recover Gemini session logs.

Continue with the canonical repository reference; this route follows the published source and does not replace it.