For any agent, task or workflow: keep context, goals, tasks, decisions, evidence and memory in one shared record with full history. Typed items hold status and acceptance criteria; agents read them in compact TOON through one CLI and SDK. Use a plain folder, a hosted workspace or sync with your team’s tools. With or without git.
Coding, research, operations, long-running workflows and recursive agents that call themselves: the same record keeps the work in context.
Install pm
npm install -g @unbrained/pm-clipm-0682l4 · Issue · P0
Recorded changes
Both changes are recorded in the item and its history.
MCP annotation and linked-artifact tools silently read when add is passed at the top level: the strict schema declares add/text but dispatch only hoists ifAbsent
priority: 1 · risk: medium
test_runs: 1 (last: passed)
comments: 2
resolution: runtime-input.ts now derives flat-to-options hoisting for comments, notes, learnings, files, docs and test from PM_TOOL_ACTION_PARAMETER_CONTRACTS, as for the read actions. Operands (id, text, query, keywords) are excluded, and nested options still win. foldAnnotationTextOperand binds the annotation text operand, flat or nested, to add, the way the CLI binds positional text. Disagreeing text and add values are refused with annotation_text_source_conflict before any write. MCP option-conformance docs describe the rule. Remaining silent drops in other action families, such as pm_history with a flat limit, are owned by pm-yjjo4l.
14 entries in .agents/pm/history/pm-wsqxwb.jsonl
Agents manage their own context, and people work alongside them. Everyone reads and updates the same typed items through the pm CLI, SDK, pm-web, remote MCP or ChatGPT.
Works with
pm context --limit 10 gives an agent the state of the project in one call: what is in progress, what is blocked, and what to pick up next. This is pm's own tracker.
summary:
returned_focus:
active_items: 10
in_progress: 0
open: 10
blocked: 2
blocked_fallback_used: false
high_level: 6
low_level: 4
agenda_events: 0
scope: "matching_items"
active_items: 247
open: 247
in_progress: 0
blocked: 64
total_items: 3236
closed: 2916
canceled: 14
high_level[6]{id,title,type,status,priority,deadline,tags,updated_at,parent,blocked,blocker_ids}:
pm-j82fd3,"Outcome milestone: agent-legible command grammar — one-screen surface, zero feature loss, alias-preserved",Milestone,open,1,"2026-10-31T00:00:00.000Z","agent-ux;grammar;horizon:near;milestone;outcome;token-efficiency","2026-10-10T19:31:13.316Z",pm-doxj,true,pm-1kxs
pm-doxj,pm-cli post-v0.1 roadmap & living backlog,Milestone,open,0,null,backlog;context-management;living-map;pm-cli;post-v0.1;roadmap,"2026-11-04T17:39:51.547Z",null,null,null
pm-rs9vry,"MCP 2026-07-28 conformance programme: stateless protocol, discovery, MRTR, subscriptions, extensions, security, caching, and released-artifact proof",Feature,open,1,"2026-12-31T00:00:00.000Z","2026-07-28;area:mcp;area:sdk;conformance;protocol;release-gate;security;stateless","2026-10-10T04:21:06.820Z",pm-5k4v,null,null
pm-f05lsg,"Recovery closure: the fraction of refusals an agent can act on using only the refusal becomes a measured, ratcheted gate",Feature,open,1,null,agent-ergonomics;ci-gates;discoverability;long-horizon,"2026-11-06T12:45:34.036Z",pm-mpbb,null,null
pm-4k6b,"Adaptive context relevance engine: multi-signal scoring over metadata, graph, and history drives context/next assembly under an explicit token budget",Feature,open,1,null,context;relevance;scale,"2026-09-22T04:59:34.890Z",pm-9rxu,true,pm-10l1vm;pm-3yzyfe;pm-atfm;pm-g8t7b9;pm-gxs55a;pm-o2kc
pm-5t33or,"Contract-declared output token budgets: every command declares a default ceiling and degrades deterministically instead of emitting unbounded output",Feature,open,1,null,"agent-ux;area:cli;area:sdk;context-management;contracts;output-contract;token-efficiency","2026-10-10T08:23:09.508Z",pm-n7rr,null,null
high_level_encoding:
json_columns: []
absent[8]:
- [2]: 1,blocked
- [2]: 1,blocker_ids
- [2]: 2,blocked
- [2]: 2,blocker_idsFive commands, in this order, for every piece of work.
See what is in progress, what is blocked and what is next.
pm context --limit 10Take one item so no other agent starts on it.
pm claim <id>Attach the files you change and the command that tests them.
pm files <id> --add src/x.ts
pm test <id> --add command="npm test"Run the linked tests. The result is recorded on the item.
pm test <id> --runClose the item with the evidence, then release your claim.
pm close <id> "<evidence>"
pm release <id>Add integrations and workflows to a project with pm install.
GitHub Issues importer extension for pm-cli
pm install npm:pm-github --projectGenerate CHANGELOG.md and release notes from completed pm items for local releases, GitHub Actions, runners, and scripts. Can run as a standalone CLI/API package or as `pm changelog generate`.
pm install npm:pm-changelog --projectGenerate deterministic pm context packs for agent handoffs, reviews, and status briefs
pm install npm:pm-context --projectToken-budgeted agent briefs and next-work plans for pm workspaces
pm install npm:pm-brief --projectKnowledge graph and dependency graph extension for pm CLI workspaces, with optional Neo4j sync.
pm install npm:pm-graph --projectAll 6 official pm-cli workspace presets in one package: bug-triage, indie-dev, open-source, software-sprint, startup-roadmap, kanban
pm install npm:pm-presets --projectThe same tracker, shared by your team in real time from the browser, ChatGPT or any MCP client.
Boards, lists and item history in the browser.
pm-web.unbrained.devRead and update your hosted pm projects from ChatGPT and Codex.
pm-gpt.unbrained.devA remote MCP endpoint for any MCP client.
pm-mcp.unbrained.devSearch the pm items of every public repository that uses pm.
pm-search.unbrained.dev