feat(routing): consolidated run summary for dispatched agents#1
Merged
Conversation
Dispatched agents are now asked to end with a structured '## What I did' section (files changed, verify result, flags), and the router prints a per-agent 'Run summary' after they finish — multi-agent runs are legible without expanding each transcript. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
RamanaNani
added a commit
that referenced
this pull request
Jun 16, 2026
* feat(routing): consolidated run summary for dispatched agents Dispatched agents are now asked to end with a structured '## What I did' section (files changed, verify result, flags), and the router prints a per-agent 'Run summary' after they finish — multi-agent runs are legible without expanding each transcript. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs: add Hina assistant design & implementation plan Corrected architecture (retrieval memory + live tools, NOT fine-tuned memory), 12 scenarios walked, the local-SLM's correct role (embeddings/style, not memory), a 3-phase roadmap, risks, and success criteria. Seeds the v1 build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): instant rating + background learning (don't block the user) Rating appends in <1s; learn.js now runs detached (&) so the bandit retrain never makes the user wait before the next task. Also documents that the installed scripts dir must contain feedback.js (stale single-script installs caused 'module not found'). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): native-style one-digit rating prompt (1/2/3/4 · 0 skip) Mirror Claude Code's inline session-feedback widget: a single compact line, reply with one digit, optional/dismissible, non-blocking. feedback.js already accepts 1-4. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * chore: keep Project Hina private — gitignore + untrack HINA_PLAN.md The Hina idea/plan stays local until it's ready to ship; not tracked or published. Also ignores a future hina/ runtime dir. File remains on disk for the build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(routing): decompose multi-task requests, fan out to multiple specialists Fixes the #1 failure mode: a bundled request ('audit X, fix SSE, add citations, optimize') was collapsed onto a single agent. Step 1a now splits it into sub-tasks, routes each to its own best specialist, dispatches independent ones in parallel and dependent ones in sequence, and shows the decomposition plan before dispatching. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
RamanaNani
added a commit
that referenced
this pull request
Jun 17, 2026
* feat(routing): consolidated run summary for dispatched agents Dispatched agents are now asked to end with a structured '## What I did' section (files changed, verify result, flags), and the router prints a per-agent 'Run summary' after they finish — multi-agent runs are legible without expanding each transcript. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs: add Hina assistant design & implementation plan Corrected architecture (retrieval memory + live tools, NOT fine-tuned memory), 12 scenarios walked, the local-SLM's correct role (embeddings/style, not memory), a 3-phase roadmap, risks, and success criteria. Seeds the v1 build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): instant rating + background learning (don't block the user) Rating appends in <1s; learn.js now runs detached (&) so the bandit retrain never makes the user wait before the next task. Also documents that the installed scripts dir must contain feedback.js (stale single-script installs caused 'module not found'). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): native-style one-digit rating prompt (1/2/3/4 · 0 skip) Mirror Claude Code's inline session-feedback widget: a single compact line, reply with one digit, optional/dismissible, non-blocking. feedback.js already accepts 1-4. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * chore: keep Project Hina private — gitignore + untrack HINA_PLAN.md The Hina idea/plan stays local until it's ready to ship; not tracked or published. Also ignores a future hina/ runtime dir. File remains on disk for the build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(routing): decompose multi-task requests, fan out to multiple specialists Fixes the #1 failure mode: a bundled request ('audit X, fix SSE, add citations, optimize') was collapsed onto a single agent. Step 1a now splits it into sub-tasks, routes each to its own best specialist, dispatches independent ones in parallel and dependent ones in sequence, and shows the decomposition plan before dispatching. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat: added the project hina v1 --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
RamanaNani
added a commit
that referenced
this pull request
Jun 17, 2026
* feat(routing): consolidated run summary for dispatched agents Dispatched agents are now asked to end with a structured '## What I did' section (files changed, verify result, flags), and the router prints a per-agent 'Run summary' after they finish — multi-agent runs are legible without expanding each transcript. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs: add Hina assistant design & implementation plan Corrected architecture (retrieval memory + live tools, NOT fine-tuned memory), 12 scenarios walked, the local-SLM's correct role (embeddings/style, not memory), a 3-phase roadmap, risks, and success criteria. Seeds the v1 build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): instant rating + background learning (don't block the user) Rating appends in <1s; learn.js now runs detached (&) so the bandit retrain never makes the user wait before the next task. Also documents that the installed scripts dir must contain feedback.js (stale single-script installs caused 'module not found'). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): native-style one-digit rating prompt (1/2/3/4 · 0 skip) Mirror Claude Code's inline session-feedback widget: a single compact line, reply with one digit, optional/dismissible, non-blocking. feedback.js already accepts 1-4. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * chore: keep Project Hina private — gitignore + untrack HINA_PLAN.md The Hina idea/plan stays local until it's ready to ship; not tracked or published. Also ignores a future hina/ runtime dir. File remains on disk for the build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(routing): decompose multi-task requests, fan out to multiple specialists Fixes the #1 failure mode: a bundled request ('audit X, fix SSE, add citations, optimize') was collapsed onto a single agent. Step 1a now splits it into sub-tasks, routes each to its own best specialist, dispatches independent ones in parallel and dependent ones in sequence, and shows the decomposition plan before dispatching. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat: added the project hina v1 * fix: updated the installation * fix: updated the installation --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
RamanaNani
added a commit
that referenced
this pull request
Jun 17, 2026
* feat(routing): consolidated run summary for dispatched agents Dispatched agents are now asked to end with a structured '## What I did' section (files changed, verify result, flags), and the router prints a per-agent 'Run summary' after they finish — multi-agent runs are legible without expanding each transcript. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs: add Hina assistant design & implementation plan Corrected architecture (retrieval memory + live tools, NOT fine-tuned memory), 12 scenarios walked, the local-SLM's correct role (embeddings/style, not memory), a 3-phase roadmap, risks, and success criteria. Seeds the v1 build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): instant rating + background learning (don't block the user) Rating appends in <1s; learn.js now runs detached (&) so the bandit retrain never makes the user wait before the next task. Also documents that the installed scripts dir must contain feedback.js (stale single-script installs caused 'module not found'). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): native-style one-digit rating prompt (1/2/3/4 · 0 skip) Mirror Claude Code's inline session-feedback widget: a single compact line, reply with one digit, optional/dismissible, non-blocking. feedback.js already accepts 1-4. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * chore: keep Project Hina private — gitignore + untrack HINA_PLAN.md The Hina idea/plan stays local until it's ready to ship; not tracked or published. Also ignores a future hina/ runtime dir. File remains on disk for the build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(routing): decompose multi-task requests, fan out to multiple specialists Fixes the #1 failure mode: a bundled request ('audit X, fix SSE, add citations, optimize') was collapsed onto a single agent. Step 1a now splits it into sub-tasks, routes each to its own best specialist, dispatches independent ones in parallel and dependent ones in sequence, and shows the decomposition plan before dispatching. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat: added the project hina v1 * fix: updated the installation * fix: updated the installation * feat(install): native plugin dependencies for one-install setup + ship Hina - plugin.json: declare 21 cross-marketplace dependencies (ecc, ruflo, wshobson@claude-code-workflows, voltagent, marketingskills) so Claude Code auto-installs the curated specialists at install time; bump to 0.4.0 - marketplace.json: allowCrossMarketplaceDependenciesOn for the 4 upstream marketplaces; bump to 0.4.0 - hina-setup.sh (SessionStart hook): now only registers the 4 upstream marketplaces so the native dependency resolver installs them; opt-in env vars for non-marketplace tools (gstack / claude-mem / claude-code-router) with their verified commands - ship skills/hina/SKILL.md (was untracked + previously gitignored, so Hina never reached main) - .gitignore: stop excluding the Hina skill Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
RamanaNani
added a commit
that referenced
this pull request
Jun 17, 2026
* feat(routing): consolidated run summary for dispatched agents Dispatched agents are now asked to end with a structured '## What I did' section (files changed, verify result, flags), and the router prints a per-agent 'Run summary' after they finish — multi-agent runs are legible without expanding each transcript. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs: add Hina assistant design & implementation plan Corrected architecture (retrieval memory + live tools, NOT fine-tuned memory), 12 scenarios walked, the local-SLM's correct role (embeddings/style, not memory), a 3-phase roadmap, risks, and success criteria. Seeds the v1 build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): instant rating + background learning (don't block the user) Rating appends in <1s; learn.js now runs detached (&) so the bandit retrain never makes the user wait before the next task. Also documents that the installed scripts dir must contain feedback.js (stale single-script installs caused 'module not found'). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): native-style one-digit rating prompt (1/2/3/4 · 0 skip) Mirror Claude Code's inline session-feedback widget: a single compact line, reply with one digit, optional/dismissible, non-blocking. feedback.js already accepts 1-4. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * chore: keep Project Hina private — gitignore + untrack HINA_PLAN.md The Hina idea/plan stays local until it's ready to ship; not tracked or published. Also ignores a future hina/ runtime dir. File remains on disk for the build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(routing): decompose multi-task requests, fan out to multiple specialists Fixes the #1 failure mode: a bundled request ('audit X, fix SSE, add citations, optimize') was collapsed onto a single agent. Step 1a now splits it into sub-tasks, routes each to its own best specialist, dispatches independent ones in parallel and dependent ones in sequence, and shows the decomposition plan before dispatching. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat: added the project hina v1 * fix: updated the installation * fix: updated the installation * feat(install): native plugin dependencies for one-install setup + ship Hina - plugin.json: declare 21 cross-marketplace dependencies (ecc, ruflo, wshobson@claude-code-workflows, voltagent, marketingskills) so Claude Code auto-installs the curated specialists at install time; bump to 0.4.0 - marketplace.json: allowCrossMarketplaceDependenciesOn for the 4 upstream marketplaces; bump to 0.4.0 - hina-setup.sh (SessionStart hook): now only registers the 4 upstream marketplaces so the native dependency resolver installs them; opt-in env vars for non-marketplace tools (gstack / claude-mem / claude-code-router) with their verified commands - ship skills/hina/SKILL.md (was untracked + previously gitignored, so Hina never reached main) - .gitignore: stop excluding the Hina skill Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(install): correct marketingskills allowlist + harden setup hook (v0.4.1) Audit (claude-code-guide + code-reviewer) confirmed native dependencies + allowCrossMarketplaceDependenciesOn is the correct mechanism, and found: - CRITICAL: marketing-skills dependency couldn't resolve — added 'marketingskills' to allowCrossMarketplaceDependenciesOn (+ its marketplace to the setup hook) - hina-setup.sh: slimmed to idempotent marketplace registration only (removed heavy backgrounded npm/git opt-in installs that could leave half-installed state); marker written early (no re-run-every-session on failure); guard node; portable date; honor CLAUDE_CONFIG_DIR; verify expected marketplace names registered (slug != name for wshobson) - hina-bootstrap.js: execSync timeout; marketplace-name fallback skips instead of guessing e.name; removed dead SETTINGS const - bump 0.4.0 -> 0.4.1 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(install): add claude-mem as a native dependency (v0.4.2) claude-mem IS a marketplace plugin (thedotmack/claude-mem) — verified by manual install — so it can be a native dependency, not npm-only as previously assumed. Added { name: claude-mem, marketplace: thedotmack } to dependencies (now 22), added 'thedotmack' to allowCrossMarketplaceDependenciesOn, and added thedotmack/claude-mem to the setup hook's marketplace list. Bump 0.4.1 -> 0.4.2. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
RamanaNani
added a commit
that referenced
this pull request
Jun 17, 2026
* feat(routing): consolidated run summary for dispatched agents Dispatched agents are now asked to end with a structured '## What I did' section (files changed, verify result, flags), and the router prints a per-agent 'Run summary' after they finish — multi-agent runs are legible without expanding each transcript. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs: add Hina assistant design & implementation plan Corrected architecture (retrieval memory + live tools, NOT fine-tuned memory), 12 scenarios walked, the local-SLM's correct role (embeddings/style, not memory), a 3-phase roadmap, risks, and success criteria. Seeds the v1 build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): instant rating + background learning (don't block the user) Rating appends in <1s; learn.js now runs detached (&) so the bandit retrain never makes the user wait before the next task. Also documents that the installed scripts dir must contain feedback.js (stale single-script installs caused 'module not found'). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): native-style one-digit rating prompt (1/2/3/4 · 0 skip) Mirror Claude Code's inline session-feedback widget: a single compact line, reply with one digit, optional/dismissible, non-blocking. feedback.js already accepts 1-4. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * chore: keep Project Hina private — gitignore + untrack HINA_PLAN.md The Hina idea/plan stays local until it's ready to ship; not tracked or published. Also ignores a future hina/ runtime dir. File remains on disk for the build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(routing): decompose multi-task requests, fan out to multiple specialists Fixes the #1 failure mode: a bundled request ('audit X, fix SSE, add citations, optimize') was collapsed onto a single agent. Step 1a now splits it into sub-tasks, routes each to its own best specialist, dispatches independent ones in parallel and dependent ones in sequence, and shows the decomposition plan before dispatching. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat: added the project hina v1 * fix: updated the installation * fix: updated the installation * feat(install): native plugin dependencies for one-install setup + ship Hina - plugin.json: declare 21 cross-marketplace dependencies (ecc, ruflo, wshobson@claude-code-workflows, voltagent, marketingskills) so Claude Code auto-installs the curated specialists at install time; bump to 0.4.0 - marketplace.json: allowCrossMarketplaceDependenciesOn for the 4 upstream marketplaces; bump to 0.4.0 - hina-setup.sh (SessionStart hook): now only registers the 4 upstream marketplaces so the native dependency resolver installs them; opt-in env vars for non-marketplace tools (gstack / claude-mem / claude-code-router) with their verified commands - ship skills/hina/SKILL.md (was untracked + previously gitignored, so Hina never reached main) - .gitignore: stop excluding the Hina skill Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(install): correct marketingskills allowlist + harden setup hook (v0.4.1) Audit (claude-code-guide + code-reviewer) confirmed native dependencies + allowCrossMarketplaceDependenciesOn is the correct mechanism, and found: - CRITICAL: marketing-skills dependency couldn't resolve — added 'marketingskills' to allowCrossMarketplaceDependenciesOn (+ its marketplace to the setup hook) - hina-setup.sh: slimmed to idempotent marketplace registration only (removed heavy backgrounded npm/git opt-in installs that could leave half-installed state); marker written early (no re-run-every-session on failure); guard node; portable date; honor CLAUDE_CONFIG_DIR; verify expected marketplace names registered (slug != name for wshobson) - hina-bootstrap.js: execSync timeout; marketplace-name fallback skips instead of guessing e.name; removed dead SETTINGS const - bump 0.4.0 -> 0.4.1 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(install): add claude-mem as a native dependency (v0.4.2) claude-mem IS a marketplace plugin (thedotmack/claude-mem) — verified by manual install — so it can be a native dependency, not npm-only as previously assumed. Added { name: claude-mem, marketplace: thedotmack } to dependencies (now 22), added 'thedotmack' to allowCrossMarketplaceDependenciesOn, and added thedotmack/claude-mem to the setup hook's marketplace list. Bump 0.4.1 -> 0.4.2. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix: updated the plan Hina --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
RamanaNani
added a commit
that referenced
this pull request
Jun 18, 2026
* feat(routing): consolidated run summary for dispatched agents Dispatched agents are now asked to end with a structured '## What I did' section (files changed, verify result, flags), and the router prints a per-agent 'Run summary' after they finish — multi-agent runs are legible without expanding each transcript. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs: add Hina assistant design & implementation plan Corrected architecture (retrieval memory + live tools, NOT fine-tuned memory), 12 scenarios walked, the local-SLM's correct role (embeddings/style, not memory), a 3-phase roadmap, risks, and success criteria. Seeds the v1 build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): instant rating + background learning (don't block the user) Rating appends in <1s; learn.js now runs detached (&) so the bandit retrain never makes the user wait before the next task. Also documents that the installed scripts dir must contain feedback.js (stale single-script installs caused 'module not found'). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): native-style one-digit rating prompt (1/2/3/4 · 0 skip) Mirror Claude Code's inline session-feedback widget: a single compact line, reply with one digit, optional/dismissible, non-blocking. feedback.js already accepts 1-4. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * chore: keep Project Hina private — gitignore + untrack HINA_PLAN.md The Hina idea/plan stays local until it's ready to ship; not tracked or published. Also ignores a future hina/ runtime dir. File remains on disk for the build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(routing): decompose multi-task requests, fan out to multiple specialists Fixes the #1 failure mode: a bundled request ('audit X, fix SSE, add citations, optimize') was collapsed onto a single agent. Step 1a now splits it into sub-tasks, routes each to its own best specialist, dispatches independent ones in parallel and dependent ones in sequence, and shows the decomposition plan before dispatching. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat: added the project hina v1 * fix: updated the installation * fix: updated the installation * feat(install): native plugin dependencies for one-install setup + ship Hina - plugin.json: declare 21 cross-marketplace dependencies (ecc, ruflo, wshobson@claude-code-workflows, voltagent, marketingskills) so Claude Code auto-installs the curated specialists at install time; bump to 0.4.0 - marketplace.json: allowCrossMarketplaceDependenciesOn for the 4 upstream marketplaces; bump to 0.4.0 - hina-setup.sh (SessionStart hook): now only registers the 4 upstream marketplaces so the native dependency resolver installs them; opt-in env vars for non-marketplace tools (gstack / claude-mem / claude-code-router) with their verified commands - ship skills/hina/SKILL.md (was untracked + previously gitignored, so Hina never reached main) - .gitignore: stop excluding the Hina skill Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(install): correct marketingskills allowlist + harden setup hook (v0.4.1) Audit (claude-code-guide + code-reviewer) confirmed native dependencies + allowCrossMarketplaceDependenciesOn is the correct mechanism, and found: - CRITICAL: marketing-skills dependency couldn't resolve — added 'marketingskills' to allowCrossMarketplaceDependenciesOn (+ its marketplace to the setup hook) - hina-setup.sh: slimmed to idempotent marketplace registration only (removed heavy backgrounded npm/git opt-in installs that could leave half-installed state); marker written early (no re-run-every-session on failure); guard node; portable date; honor CLAUDE_CONFIG_DIR; verify expected marketplace names registered (slug != name for wshobson) - hina-bootstrap.js: execSync timeout; marketplace-name fallback skips instead of guessing e.name; removed dead SETTINGS const - bump 0.4.0 -> 0.4.1 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(install): add claude-mem as a native dependency (v0.4.2) claude-mem IS a marketplace plugin (thedotmack/claude-mem) — verified by manual install — so it can be a native dependency, not npm-only as previously assumed. Added { name: claude-mem, marketplace: thedotmack } to dependencies (now 22), added 'thedotmack' to allowCrossMarketplaceDependenciesOn, and added thedotmack/claude-mem to the setup hook's marketplace list. Bump 0.4.1 -> 0.4.2. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix: updated the plan Hina * feat(hina): self-contained engineering team + gated build pipeline + hybrid memory (v0.5.0) Turn the roster into an enforced engineering org Hina conducts: - 10 baseline agents so Hina builds end-to-end with zero external plugins: requirements-analyst, ux-visualizer, delivery-orchestrator, solution-architect, frontend/backend/data-engineer, build-fixer, security-engineer, qa-verifier. All registered in data/registry.json; all carry the 4 dispatch guardrails. - workflows/build-pipeline.md: enforced flow with owners, handoff contracts, and two HARD gates (TEST: qa-verifier; VALIDATE: code-reviewer + security-engineer) + loop-back. - delivery-orchestrator emits a concrete gated task-DAG. - hina-memory.js: hybrid `recall` (lexical + dense cosine, RRF-fused) replacing tail-only memory; observe stores embeddings. Wired into Hina loop step 0 (observe step). - TEAM.md documents the roster, hierarchy, and pipeline. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
RamanaNani
added a commit
that referenced
this pull request
Jun 18, 2026
* feat(routing): consolidated run summary for dispatched agents Dispatched agents are now asked to end with a structured '## What I did' section (files changed, verify result, flags), and the router prints a per-agent 'Run summary' after they finish — multi-agent runs are legible without expanding each transcript. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs: add Hina assistant design & implementation plan Corrected architecture (retrieval memory + live tools, NOT fine-tuned memory), 12 scenarios walked, the local-SLM's correct role (embeddings/style, not memory), a 3-phase roadmap, risks, and success criteria. Seeds the v1 build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): instant rating + background learning (don't block the user) Rating appends in <1s; learn.js now runs detached (&) so the bandit retrain never makes the user wait before the next task. Also documents that the installed scripts dir must contain feedback.js (stale single-script installs caused 'module not found'). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(feedback): native-style one-digit rating prompt (1/2/3/4 · 0 skip) Mirror Claude Code's inline session-feedback widget: a single compact line, reply with one digit, optional/dismissible, non-blocking. feedback.js already accepts 1-4. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * chore: keep Project Hina private — gitignore + untrack HINA_PLAN.md The Hina idea/plan stays local until it's ready to ship; not tracked or published. Also ignores a future hina/ runtime dir. File remains on disk for the build. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(routing): decompose multi-task requests, fan out to multiple specialists Fixes the #1 failure mode: a bundled request ('audit X, fix SSE, add citations, optimize') was collapsed onto a single agent. Step 1a now splits it into sub-tasks, routes each to its own best specialist, dispatches independent ones in parallel and dependent ones in sequence, and shows the decomposition plan before dispatching. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat: added the project hina v1 * fix: updated the installation * fix: updated the installation * feat(install): native plugin dependencies for one-install setup + ship Hina - plugin.json: declare 21 cross-marketplace dependencies (ecc, ruflo, wshobson@claude-code-workflows, voltagent, marketingskills) so Claude Code auto-installs the curated specialists at install time; bump to 0.4.0 - marketplace.json: allowCrossMarketplaceDependenciesOn for the 4 upstream marketplaces; bump to 0.4.0 - hina-setup.sh (SessionStart hook): now only registers the 4 upstream marketplaces so the native dependency resolver installs them; opt-in env vars for non-marketplace tools (gstack / claude-mem / claude-code-router) with their verified commands - ship skills/hina/SKILL.md (was untracked + previously gitignored, so Hina never reached main) - .gitignore: stop excluding the Hina skill Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(install): correct marketingskills allowlist + harden setup hook (v0.4.1) Audit (claude-code-guide + code-reviewer) confirmed native dependencies + allowCrossMarketplaceDependenciesOn is the correct mechanism, and found: - CRITICAL: marketing-skills dependency couldn't resolve — added 'marketingskills' to allowCrossMarketplaceDependenciesOn (+ its marketplace to the setup hook) - hina-setup.sh: slimmed to idempotent marketplace registration only (removed heavy backgrounded npm/git opt-in installs that could leave half-installed state); marker written early (no re-run-every-session on failure); guard node; portable date; honor CLAUDE_CONFIG_DIR; verify expected marketplace names registered (slug != name for wshobson) - hina-bootstrap.js: execSync timeout; marketplace-name fallback skips instead of guessing e.name; removed dead SETTINGS const - bump 0.4.0 -> 0.4.1 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(install): add claude-mem as a native dependency (v0.4.2) claude-mem IS a marketplace plugin (thedotmack/claude-mem) — verified by manual install — so it can be a native dependency, not npm-only as previously assumed. Added { name: claude-mem, marketplace: thedotmack } to dependencies (now 22), added 'thedotmack' to allowCrossMarketplaceDependenciesOn, and added thedotmack/claude-mem to the setup hook's marketplace list. Bump 0.4.1 -> 0.4.2. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix: updated the plan Hina * feat(hina): self-contained engineering team + gated build pipeline + hybrid memory (v0.5.0) Turn the roster into an enforced engineering org Hina conducts: - 10 baseline agents so Hina builds end-to-end with zero external plugins: requirements-analyst, ux-visualizer, delivery-orchestrator, solution-architect, frontend/backend/data-engineer, build-fixer, security-engineer, qa-verifier. All registered in data/registry.json; all carry the 4 dispatch guardrails. - workflows/build-pipeline.md: enforced flow with owners, handoff contracts, and two HARD gates (TEST: qa-verifier; VALIDATE: code-reviewer + security-engineer) + loop-back. - delivery-orchestrator emits a concrete gated task-DAG. - hina-memory.js: hybrid `recall` (lexical + dense cosine, RRF-fused) replacing tail-only memory; observe stores embeddings. Wired into Hina loop step 0 (observe step). - TEAM.md documents the roster, hierarchy, and pipeline. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(plugin): drop hard marketing-skills dependency that blocked loading (v0.5.1) An unmet hard dependency (marketing-skills@marketingskills, not installed) made Claude Code refuse to load the whole plugin, hiding /hina and agent-router. The router discovers tools at runtime and falls back to its own baselines, so it must not force-install anything (HINA_PLAN.md §11 — propose-then-install, never silent). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Dispatched agents are now asked to end with a structured '## What I did' section (files changed, verify result, flags), and the router prints a per-agent 'Run summary' after they finish — multi-agent runs are legible without expanding each transcript.
What does this PR do?
Type
Checklist
node --check scripts/*.jspassesdata/skills-index.json, logs, learned overlay stay out — see CONTRIBUTING)CHANGELOG.mdif behavior changedNotes for the reviewer