agentsys roadmap

Curated capabilities beyond vendor connectors. Status: planned (not started), in progress (issue or PR open), shipped (on main).

Orchestration features below use optional keys in .agentsys/profile.json (routing, review, templates).

Highest value

Item	Why	Effort	Risk	Status
Create and edit tasks in the dashboard	Start work without the CLI	Low	Low	planned
Dispatch from the UI (adapter/account picker, budget preview, confirm)	Most-requested control surface	Medium	Spends real quota	planned
Notifications (desktop, Slack) for approvals and escalations	Reduces time waiting on humans	Low	Low	planned
Spend in dollars (user-maintained price table)	Token caps mislead when cost varies by model	Medium	Prices drift	planned
Per-model limits and allow-lists per account	Prevent expensive model surprises	Medium	Medium	planned

Item	Why	Effort	Risk	Status	Tracking
Routing rules (`--adapter auto`, `agentsys route`)	Use several subscriptions efficiently within each account's caps	Medium	Must not bypass caps	shipped	#9, PR #15
Review assignment (`review.pairs`, `agentsys reviewer`)	Independent review by harness, not luck	Low	Low	shipped	#10, PR #16
Task templates and pipelines (`agentsys pipeline`)	Encode research → implement chains	Medium	Low	shipped	#11, PR #17
Scheduler / watch mode (`agentsys watch`)	Optional background dispatch when rules exist	Medium	Spends quota if misconfigured	shipped	#12, PR #18

Item	Why	Effort	Risk	Status
Deeper verification (auto-detect test/lint from repo)	Stronger gate without manual `--check` every time	Medium	False positives	planned
Signed identities (bind actors to CLI login or local key)	Fixes self-declared actor trust gap	High	High	planned
Adapter health checks (`adapters --check`, doctor)	Vendor JSON breaks without notice	Low	Live check spends tokens	shipped (#13, PR #19)

Item	Why	Effort	Risk	Status
Agent scorecards (pass rate, tokens per accepted task)	Compare vendors on real outcomes	Medium	Low	planned
Import IDE session usage (extend `usage_import`)	Dashboard under-reports interactive chat	Medium	Format churn	planned
Exportable audit reports (per task or sprint)	Compliance and postmortems	Low	Low	planned

Item	Why	Effort	Risk	Status
PyPI release and Homebrew formula	Easier install than git clone	Low	Packaging maintenance	planned
GitHub integration (PR from accepted task, check comments)	Fits existing git workflow	High	API scope	planned
Team mode (shared store over git or small server)	Multi-person coordination	High	Product shape change	planned
Demo video / GIF on landing page	Faster comprehension	Low	Low	planned

Item	Status	Tracking
Grok Build (`grok`)	shipped	`main`
Gemini CLI + Antigravity (MCP) coordination	planned	#6
OpenCode (`opencode run`)	shipped	#7

Item	Status	Tracking
This roadmap	shipped	#8, PR #14

See also PRODUCT_BRIEF.md §7–8 for limitations and owner questions that informed this list.