| name | team-worklog |
| description | Use at the start of any non-trivial task in a multi-agent/multi-engine setup — to see who did what, at what stage, and what's still open or broken — and throughout the work, to record steps, failures, and results, so any other agent or engine can pick up the context. Триггеры (RU) — «worklog», «летопись», «журнал работ», «что делалось / на какой стадии», перед началом задачи, при сбое, при сдаче. Cross-engine shared journal for teams running several AI engines/agents against the same codebase or project. |
| license | MIT |
Team worklog — a shared append-only ledger across engines
A single append-only event log: who, when, what was done, what failed, what came out of
it, what stage things are at right now. Append-only: never edit or delete another
engine's entries — only add your own.
The mechanism used here is the ledger command of samemind
(npx samemind ledger …) — a small, git-native, dependency-light memory tool. You don't
need to build your own journal script: samemind ships this ledger as one of its
commands, and it's the same public package this skill's author uses in production. If
your team already has its own shared-journal mechanism, adapt the steps below to it —
the discipline (read first, log meaningful steps, always log failures) is what matters,
not the specific CLI.
1. ENTRY — read before you start work
Before picking up a task, pull the current context:
npx samemind ledger status
npx samemind ledger read --topic <topic-id> --limit 20
Draw conclusions: don't redo what's already done; check for open failures on your topic before starting.
2. DURING WORK — log meaningful steps
One entry per meaningful step/decision/failure — not per tool call (that's noise), but at
the points that matter: started, a key decision, an intermediate result, a failure, handed off.
npx samemind ledger append \
--actor <your-engine-or-agent-id> \
--topic <stable-id: branch / ticket / task slug> \
--phase <start|step|done|fail|block|note> \
--status <ok|wip|partial|fail> \
--action "what exactly was done/decided/broke — fold in the next step here if there is one"
[--artifact <branch/commit/file>] [--ref <issue/link>]
3. EXIT — record the outcome
- Success:
--phase done --status ok (+ --artifact with the commit/branch).
- Failure/blocked:
--phase fail (or block) --status fail --action "what broke — and what's needed next". Always log a failure — otherwise it's invisible to the next agent.
Rules
topic is a stable identifier (one branch/task = one topic id), otherwise events won't group into a coherent timeline.
actor is your own engine/agent name — that's how anyone reading the ledger later can tell who did what.
- A failure with no entry is a lost failure. Always log
fail.
- Don't invent progress — write only what actually happened.
- Never edit someone else's entries — only append your own.
Examples
npx samemind ledger status
npx samemind ledger append --actor grok --topic auth-refactor --phase start --status wip \
--action "picked up the task: refactor authorization"
npx samemind ledger append --actor grok --topic auth-refactor --phase fail --status fail \
--action "tests fail on refresh-token — need to fix the TTL and re-run"
npx samemind ledger append --actor grok --topic auth-refactor --phase done --status ok \
--action "authorization refactored, tests green" --artifact "auth-refactor@a1b2c3d"
Why a ledger instead of ad hoc notes
Without a shared ledger: every engine keeps its own private sense of "what's done", two
agents redo the same work, and a failure one engine hit gets silently repeated by the
next one that picks up the same topic. The ledger is the one place any engine — human or
AI, on any host — can check first and append to last.