xiaobright/dsh-anchored-standard

Two-phase DeepSeek Harness preset: Minimal-aligned bootstrap, then full Standard tools (Project2 98/99)

View on GitHub ↗Jump to charts ↓Open shareable report

Summary Information

Updated 49 minutes ago
Added to GitGenius on September 18th, 2026
Created on August 14th, 2026
Open Issues & Pull Requests: 21 (+3)
GitHub issues: Enabled
Number of forks: 120
Total Stargazers: 3,806 (+0)
Total Subscribers: 4 (+0)

Repository Insights (GitGenius)

Median issue/PR response: 11.9 hours
Mean response time: 18.7 hours
90th percentile: 39.5 hours
Tracked items: 46

Most active contributors

Sign in to see contributor activity.

How this project is maintained

About 4% of issues opened in the past year have never received a reply. 63% of issues opened in the past year have been closed, leaving a working backlog. Three people close 74% of everything that gets resolved.

Charts & Analytics

Fetching additional details & charts...

Issue Activity (beta)

Open issues: 18
New in 7 days: 0
Closed in 7 days: 0
Avg open age: 31 days
Stale 30+ days: 12
Stale 90+ days: 0

Recent activity

Opened in 7 days: 0
Closed in 7 days: 0
Comments in 7 days: 0
Events in 7 days: 0

Top labels

No label distribution available yet.

Most active issues this week

No issue events were indexed in the last 7 days.

Detailed Description

dsh-anchored-standard is a DeepSeek Harness agent preset that implements a two-phase initialization strategy for language model sessions.

The tool addresses the problem of model trajectory instability by anchoring sessions on minimal conditions before promoting to fuller tool catalogs. It works by starting with a real Minimal tool schema without auto-injected context, then transitioning to a small resident catalog once the session proves durable, which unlocks access to heavier Standard tools on demand. The preset includes a base mode, two live-anchor variants, and one seeded prefab mode, along with unified injection control through a context-gate mechanism.

Developers working with DeepSeek models who need to stabilize agent behavior across tool-rich environments should consider this preset. It suits projects where session durability matters and where tool availability can be staged rather than provided all at once. The tool is explicitly a community project not affiliated with or endorsed by DeepSeek.

The project is frozen as of September 2026 following the retirement of the DeepSeek V4 Pro model that the presets were tuned for. The maintainer halted active development earlier due to cost constraints on the evaluation loops required for preset refinement, moving the repository into maintenance-only mode. The repository remains online with its history, data, and acknowledgement record intact, and the maintainer has published a closing note explaining the decision. Research material and model-agnostic components like the context-gate injection control and dose-response discovery data remain available for reference.