I fix other people's projects, and I maintain a plugin family for DeepSeek Harness. Independent open-source developer in Beijing — not affiliated with DeepSeek, and not a DeepSeek employee.

Upstream — 416 merged pull requests in 80 repositories that are not mine, from AWS, Apple, Docker, Microsoft, NVIDIA, Redis, Tencent and the vLLM project down to one-person projects. The family — 91 actively maintained Apache-2.0 plugins in a 103-repo family: security, workflows, research, messaging bridges, developer experience, and since 2026-10-08 a set of 53 compliance checkers that read a register or record set against a versioned rule pack and report the differences. Every plugin has one page at perrylink.github.io — 114 pages, complete in the served HTML with no JavaScript rendering them, so a reader who arrives from a search result sees the text rather than a loading state.

Separate from the harness: I train small language models. Phocinae-Largha-150M-v1 (Apache-2.0) is a typed decision model — no text generation; one forward pass returns a verdict per question with calibrated confidence. GPU 21.0 ms (RTX 5090) p50 per decision, CPU-only 1.64 s per case; English typed-decisions 0.906 (400 cases / 2,000 decisions, re-measured 2026-10-09), Chinese (machine-translated eval set) 0.848. With a τ=0.6 escalate gate, 55.0% of agent decisions stay local (79.6% at τ=0.5) and kept-subset accuracy moves 0.906 → 0.9936 (+0.0876) — the gate makes the system better, not just cheaper. Weights on Hugging Face and ModelScope; server pip install phocinae-server, DSH bundle npm i dsh-phocinae. Reproduction ships with the repo: seeds, row-set hashes, environment and eval scripts.

The family publishes public evidence and scores instead of another star ranking:

• dsh-score — per-surface evidence registry, badges, bench runner and contribution receipts. Live: https://perrylink.github.io/dsh-attention-index/. Prior-only rows are E0 / unrated; paired with / without bench results are the only path to a quality slot. Repo rename to dsh-score-index pending.
• dsh-community-bench — the in-app consumer: Settings → Plugins → Community Bench. It reads the public registry; it does not compute a local score.

dsh plugin --profile web add dsh-community-bench

Every repository below is external to PerryLink/*, every row is merged work, and open proposals are deliberately not listed. ★ re-measured 2026-10-19 for every row; the third column names the party that owns the project, because the account name alone does not say whether that is a company, a standards body or one person. Every external repository above 2,500★ that carries merged work is listed.

Seven more repositories between 1,000★ and 2,500★, the long tail below that, the per-merge record behind the 416, the laya ledger and the deepseek-ai picture are all in PROJECTS.md.

99 repositories declare the dsh.bundle.patch contract and 91 are actively maintained (3 retired, 5 frozen), plus the support repositories — dsh-kit, dsh-catalog, the certification registry and the CI checker — and one third-party plugin carried as a row. The roster's source of truth is plugins.txt in dsh-kit, with a parity script holding the family to that file; the machine-readable catalogue is dsh-catalog, which is also the DSH Desktop Market source. One line per plugin, with its status and npm name, is in PROJECTS.md — 94 rows in ten tables.

A maintainer review against the official harness and the wider ecosystem retired three plugins because the harness now implements the capability natively — dsh-background-agents (native continuable subagents), dsh-session-pin (built-in session pinning) and dsh-team-rooms (Agent Teams) — and froze six because a better-adopted alternative now exists: dsh-budget, dsh-draw, dsh-claude-move, dsh-reach, dsh-wechat and dsh-defend.

Retired does not mean broken: compatibility updates stop, existing installs keep working, and the capability is either official now or served better elsewhere. The per-plugin measurements — this repository's weekly downloads against the alternative's, and what each row was retired in favour of — are in PROJECTS.md, under "Retired & frozen — the full tables".

• GitHub (this profile), Gitee and npm — source, CI and releases here; 99 family repos mirrored to Gitee by a daily job (default branch and all tags). The perrylink npm account holds 107 names and 1,601 versions; 95 of them carry a SLSA provenance attestation on at least one version, 75 on the current latest.
• npm downloads — 228,164 over the trailing 30 days (window 2026-09-19..10-18). dshfind indexes 106 of the family's repositories and publishes a download badge for 9 of them — three at 10k+, two at 5k+, one at 2k+, two at 1k+ and one at 500+ — which add to 44,500+. The badges are rounded tiers, so that is a floor and not a total.
• DSH Desktop Market — add https://cdn.jsdelivr.net/gh/PerryLink/dsh-catalog@main/deploy/artifacts/v1/plugins.json under Market → Sources to browse the family in-app (99 packages, catalogue generated 2026-10-18).
• MCP Registry — three servers, all published from their release workflows over GitHub OIDC: dsh-cert-mcp, jevcore-mcp and laya-mcp.
• GitHub Actions — dsh-github and dsh-test-drive also ship composite actions, so they install as uses: PerryLink/dsh-test-drive@vX.
• Third-party directories — a dozen-plus DSH lists and indexes ingest the dsh-plugin topic; the family is scored on OpenSSF Scorecard.

Measured, not remembered. The full inventory — the merge-by-merge upstream record, the 94-row plugin roster, the retired and frozen tables, and the npm and registry figures with their caveats — is in PROJECTS.md. How every figure is measured is in INSTRUMENTATION.md, and what each round corrected is in CHANGELOG.md.