v0.1.3

dsh-subagent-rules 0.1.3 - anchored-safe rules injection

• rules message is now gated on PROMOTION: it lands only after the session's first durable tool/call or assistant/message, so anchored-standard's request #1 stays Minimal-exact (clean tool pair, no injected context)
• verified: fresh session -> no injection; promoted session -> injected once; assistant/message also promotes; dedupe holds

v0.1.2

dsh-subagent-rules 0.1.2 - rules now reach complete-persona presets

• rules injected as a per-step message (source.kind: subagent-rules) instead of a system-prompt section: sections are collapsed to the persona by complete:true presets (e.g. anchored-standard), messages are not
• injected once per session (in-process dedupe); kind is not in anchored's suppressedContextSources, so it survives the bootstrap strip
• inject list trimmed to [tools, subagents]

v0.1.1

dsh-subagent-rules 0.1.1 - hardening release

• guard empty request configs in the effort floor listener
• effort_set validates values up front (max/high/medium/low/minimal/auto)
• tolerate same-name tool registrations from nearer scopes (anchored preset rows)
• README: compatibility notes (subagents dependency, scope shadowing, injectRules trade-off), official dsh-plugin topic guidance

v0.1.0

dsh-subagent-rules 0.1.0 - 子代理模型与思考强度规定 (subagent model & thinking-effort rules)

• subagent_flash tool: subagent pinned to the flash model route (opencode-go/deepseek-v4-flash)
• thinking effort: defaults to max on every request (fixes subagent routes not inheriting effort) + effort_set tool for per-session override
• dispatch rules auto-injected into every conversation (systemPrompt section)
• zero-dependency hand-written ESM, no build step