DSH HUB
HomePlugin StorePlugin PacksCommunityRankingsResourcesPublish Guide
Plugin source
Back to catalog

fengshaoA /

fengshaoA/dsh-skills-fengshaoA

Topic repository only

Agent skills for DeepSeek Harness, installable via skills.sh. Includes decision-council: structured deliberation with a three-tier safety gate.

★ 1 Stars0 Forks0 IssuesN/A Community rating0 Confirmed installs
View on GitHub
READMESource: main@bc2846c0

dsh-skills-fengshaoA

Agent skills for DeepSeek Harness, installable via skills.sh.

Author: 风少 (fengshaoA)

Skills

Skill What it does
decision-council Structured deliberation for consequential or irreversible decisions. 30 portable mechanisms distilled from Chinese, Western and global staff institutions, a six-step council process, and a three-tier safety gate.

Install

npx skills add fengshaoA/dsh-skills-fengshaoA

Or one skill only:

npx skills add fengshaoA/dsh-skills-fengshaoA -s decision-council

decision-council

Most bad decisions do not fail for lack of information. They fail because nobody in the process was allowed to say no.

This skill reduces real staff institutions — the Tang send-back review, the dual-key draft-and-approve, the tribune's veto, mission-type command, net assessment, scenario planning, red teams — into 30 actions you can run today, then wraps them in a six-step process: frame → grade the facts → seat a heterogeneous council → independent testimony and cross-examination → premortem and red team → weighted convergence with an eight-point audit.

It refuses to be a celebrity oracle. No "Confucius says". Each seat is an institutionalised way of doubting, not a persona.

Safety gate

Three questions before anything convenes:

  • L1 — immediate crisis. Self-harm or suicidal thinking: stop the deliberation, ask directly, give the hotline. Explicitly forbidden to run a cost-benefit analysis on staying alive.
  • L2 — hard refusal. Crimes, evading regulators, harming a specific person, rigging a case to deceive a third party.
  • L3 — restricted mode. Life-and-death, medical, legal decisions that are not emergencies: do not refuse. Refusing pushes the person back into isolation. Instead, restrict the scope — map the options, elicit what the person values, generate the questions for their doctor or lawyer — and never predict survival or cure rates, never decide for them whether to treat, never convert a life into a cost-benefit ratio.

The L3 tier is the least common design. Most safety policies refuse anything touching life and death; the result is that the people who most need help get turned away.

Provenance

The 30 mechanisms trace back to a full lineage index of 178 entries across Chinese, Western and global advisory institutions. Each entry in [1]–[30] carries its source URL and confidence level, so a reader can spot-check any claim without leaving the skill file. The selection rationale — what was kept, what was cut, and why — is in SELECTION.md.

License

Apache-2.0

—/ 5

No ratings yet

Manifest verification required

Commit bc2846c0cfdc

Community comments

No comments yet. Be the first to write one.

DSH HUB

A community index for DSH plugins. Not an official GitHub or DeepSeek AI product.

CommunityResourcesAPIAbout