DeepSeek Harness Plugin

mkpoli/dsh-history-search

Category Tools & Capabilities Added 2026-11-10

Adds one agent tool (read-only on your transcripts) that searches Claude Code, Codex and DSH session transcripts and returns each match with its resume command. Needs ripgrep (rg) and zstd (zstdcat) on PATH and keeps a plain-text cache of DSH sessions under ~/.dsh/cache/history-search.

Install

# from a prebuilt release tarball

dsh plugin --profile web add "https://github.com/mkpoli/dsh-history-search/releases/download/v0.1.0/dsh-history-search.tgz"

# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)

dsh plugin --profile web add github:mkpoli/dsh-history-search

Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).

README

A DeepSeek Harness plugin that gives the agent one search tool over the transcript stores of three CLI agents:

Source Store
Claude Code ~/.claude/projects/**/*.jsonl
Codex ~/.codex/sessions/**/rollout-*.jsonl
DeepSeek Harness ~/.dsh/sessions/**/session.*.jsonl*

The tool is read-only. It returns the newest matching sessions with excerpts, working directories, and the command that resumes each session (claude --resume, codex resume, dsh tui --resume).

Install

dsh plugin --profile web add github:mkpoli/dsh-history-search

Restart the host afterwards. The plugin declares its bundle patch in cordis.patch.yml, so mounting it adds one history-search row with default configuration.

Use

Ask the agent to search earlier work, for example:

Search my history for the spectrogram rendering fix.

The model calls history_search with:

Argument Meaning
query Literal words; every word must appear in the same transcript record.
source all (default), claude, codex, or dsh.
project Case-insensitive substring of the session working directory.
limit 1–30 sessions, newest first. Default 8.

Matching is case-insensitive. A multi-word query requires all words on one record, which keeps results focused on messages that discuss the whole phrase.

How it reads each store

Claude Code and Codex are searched in place with ripgrep, so results always reflect the current files.

DeepSeek Harness appends one zstd frame per batch to each session log. Node's zstdDecompressSync stops after the first frame, so the plugin decompresses each log once with zstdcat into ~/.dsh/cache/history-search/dsh/ and refreshes an entry only when its size or mtime changes. The cache holds plain JSONL, is created with owner-only permissions, and mirrors live sessions as they are written. Delete ~/.dsh/cache/history-search to reclaim the space; the next search rebuilds it (about 10 seconds for a few hundred sessions).

Requirements

  • ripgrep (rg) on the server's PATH.
  • zstdcat from the zstd package on the server's PATH.

Both are reported as tool errors when missing.

Limits

  • Searches record text found in messages, reasoning, tool input, and tool output. Base64 attachments and other binary payloads are skipped.
  • Subagent sidechains are included; their resume command points at the parent session.
  • Older DeepSeek Harness sessions keep their historical id, which may carry a session- prefix.

Content from the project README on GitHub ↗

Links

More in this category

View the whole category →

Community comments

Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.