Skip to content

feat: agent lifecycle, integrations and shared context (v0.0.7) - #14

Merged
codejunction merged 40 commits into
mainfrom
feature/agent-integrations
Oct 4, 2026
Merged

codejunction merged 40 commits into
mainfrom
feature/agent-integrations

Conversation

@codejunction

Copy link
Copy Markdown
Owner

What and why

Release 0.0.7: herdr/dotpals-style agent tracking plus persistent shared context.

  • Server-owned agent lifecycle (integration > custom > native > manifest > heuristic), POST /api/agent-reports with per-terminal tokens, sd agent status/wait/report/explain.
  • Hook integrations: Claude Code, Codex, Gemini CLI, Devin CLI verified; 15 more as preview (sd integration install).
  • Native launch flags, native slash commands (sd agent cmd, Actions menu), native resume (agent_resume setting).
  • SSE /api/events, sd agent prompt, versioned detection rules with local overrides.
  • Attention UI: rollups, Open next, done-until-seen; decluttered sidebar.
  • Context store (context.db, FTS5): task/state/facts/decisions, .shelldeck/ projections, deterministic capture from hooks, agent-driven fact extraction, smart recall via an agent's small model (no embeddings), sd init.
  • Version bump to 0.0.7, CHANGELOG section, refreshed README screenshots, restructured docs/AGENT_ENHANCEMENT.md.

Released as 0.0.7rc1–rc4 on PyPI from this branch first.

Checklist

  • uv run ruff check and uv run pytest -q pass (139 passed, 1 skipped)
  • Tests added or updated for behaviour changes
  • UI changes checked in a real browser (screenshots in docs/assets/)
  • README.md and CHANGELOG.md updated

🤖 Generated with Claude Code

https://claude.ai/code/session_01PobiKTh2wyjWp5KigmW5Fa


Generated by Claude Code

codejunction and others added 30 commits October 4, 2026 02:49
- agent_state registry + report validation; heuristic is fallback
- POST /api/agent-reports with per-terminal token, /api/agent-status, /api/agent-wait
- context.py: tasks/state/memory/decisions/events, fts5 recall, stale, redaction, md projection
- sd agent status|wait|report, sd context|recall|memory|remember|discover|decide|task update|knowledge verify
- update task.md
- sd context at start, recall/remember/decide during, task update before stop
- update task.md
- readme, changelog, claude.md, agent-automation.md, report schema
- update task.md
…ot, opencode

- python -m shelldeck.hook maps agent hook events to lifecycle reports
- merged, backed-up, atomic installers; only shelldeck entries touched; outdated status
- report token alone accepted from localhost only
- sd integration detect|status|install|uninstall, /api/integrations/{agent}, agents page section
- docs/agent-integrations.md; update task.md
- codex rollout events -> native working/done + codex resume <id>
- claude sessions/<pid>.json busy + claude --resume <id>
- native reports expire fast so prompts fall back to screen
- update task.md
- agent_resume setting never|ask|auto (default ask)
- resumable sessions table, sd agent resume, POST /api/sessions/{id}/resume
- argv revalidated, plain words only; skips open handoffs, missing cwd/cli
- update task.md
…, devin

- codex hooks.json + [features] hooks = true (herdr format)
- session-only reports store resumable session, no lifecycle claim
- resume argv for codex, copilot, cursor, devin, droid, qwen, qodercli
- install needs agent config folder; bulk install skips missing, reports each
- update task.md
- kimi marked toml hooks block; askuserquestion -> blocked question
- mastracode flat hooks, antigravity named block, grok own file, kilo plugin
- hook accepts explicit state arg like herdr
- 15 installable agents; update task.md
- letta default conversation resumes per agent
- hermes python plugin + plugins.enabled yaml edit, refuses unknown layouts
- 17 installable agents; update task.md
- shared extension template, herdr event names and tui gates
- node-driven tests for opencode, kilo, pi, omp plugins
- 19 installable agents; update task.md
- devin full lifecycle from hooks, session from sessions.db fallback
- gemini resume argv
- priority tier; others marked later/preview, out of default install
- update task.md
- sidebar rollup needs you > working > done, text not color
- agents sorted by urgency, elapsed time, open next across projects
- done stays until terminal opened; new tabs get current states
- update task.md
- deciding source, report ages, heuristic inputs, integration hint
- --file checks saved screen against question rules; never shows screen text
- update task.md
- POST /api/agent-prompt refuses blocked agents, waits for a state after the prompt
- SSE /api/events with last-event-id; sd events subscribe
- sd agent rename; update task.md
- team.launch_line: model, prompt flag, per-run hooks
- claude --settings per-run hooks file, skipped when global hook installed
- POST /api/agent-start, sd agent start; page launch server-side
- update task.md
- agent_commands: one action name -> claude/codex/gemini/devin native command
- sd agent cmd, POST /api/sessions/{id}/agent-command, command menu on agents page
- codex hooks add tool use, permission request, precompact (from codex source)
- gemini --resume confirmed from docs; devin marked unverified (devin.ai blocked)
- update task.md
- task/state editor, memory verify/invalidate, decisions, relations, recall search
- local-only note; not polled so edits stay
- update task.md
- agent_detection/default.toml, detection.py load/match
- per-agent overrides replace/disable/add; invalid ones ignored and reported
- explain shows rule id and manifest; docs/agent-detection.md
- update task.md
- sd switch checkpoints and starts another agent from sd context
- message button on agents page, refused while blocked
- skill teaches agent prompt/cmd/wait/switch
- drop claude StopFailure hook (not in claude 2.1.289)
- update task.md
- POST /api/sessions/{id}/extract, sd agent extract, extract facts button
- short prompt, no memory read needed (server dedups/redacts)
- sd remember takes several facts per call
- update task.md
- tier, method, exact state, resume, verified-against per agent
- changelog links it; update task.md
- stale screen 'blocked' could win a timestamp tie on windows (16ms clock)
- update task.md
… sessions

- hooks attach finished tool call (command+ok or file), never output
- events, tests line, last error; session md per agent run on exit
- people's shell commands captured too; zero tokens
- fix: redact per value so event json can't break
- update task.md
…rces

- devin slash commands expanded and verified; exec shell tool captured
- test guards devin hook events (devin rejects unknown ones)
- update task.md
claude added 10 commits October 4, 2026 19:16
- recall_agent setting, sd recall --smart, context page toggle
- claude -p haiku / gemini flash-lite / codex exec / devin -p small, query only
- keywords parsed and cached; falls back to plain search
- update task.md
- terminals, ai agents, context always shown; other views under a remembered More fold
- settings on its own line; context gets layers icon
- update task.md
…search

- working-on summary with edit on demand, compact facts/decisions, plain-words activity
- search widens with chosen installed agent + model (cheapest default)
- sd recall --agent --model, GET /api/context/recall-agents
- update task.md
- adds project if new, writes STATE/TASK/MEMORY/DECISIONS, AGENTS.md, .gitignore
- AGENTS.md never overwrites user edits; --task records first task; idempotent
- update task.md
@codejunction
codejunction merged commit 89c4663 into main Oct 4, 2026
10 checks passed
@codejunction
codejunction deleted the feature/agent-integrations branch October 4, 2026 20:07
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants