edisyl action

Every ask, teach, and task is an Action: a durable id, typed output, and an event history you can follow from the CLI.

edisyl action get <action-id>               # current status + typed output + plan reference
edisyl action plan <action-id>              # the plan behind the work + each step's progress
edisyl action step <action-id> <step-id>    # one step's brief and complete result
edisyl action watch <action-id>             # follow the work; result on stdout, narration on stderr
edisyl action events <action-id>            # raw event tail, for diagnosis
edisyl action cancel <action-id>            # request cancellation
edisyl action trace <action-id>             # content-record ids produced by a teach, for `forget`

action watch is the follow experience

Never write a polling loop. watch narrates only transitions: the plan when it appears, each step as it starts and closes with its one-line summary, anything blocked, and the typed output when the Action settles.

• Narration goes to stderr; the result goes to stdout: edisyl action watch <id> > result.txt keeps the result alone.
• Under -j, narration becomes newline-delimited JSON progress objects on stdout instead.
• Exits 0 on succeeded, declined, or waiting. Exits non-zero on failed or canceled.

Terminal statuses

Status	Meaning
succeeded	Carries the typed output.
declined	A deliberate refusal: the pack understood the request and couldn’t or wouldn’t act, with the reason in the Action’s error message. Distinct from failed.
failed	Something broke.
canceled	Someone asked it to stop.
waiting	The work stopped because it needs something from you. waitingOn.detail is the question, in the words the verb asked it.

Nothing resumes a waiting Action; answer what it names and start a new one:

edisyl ask growth-metrics "How many active wallets last week?" --wait
#   Status:  waiting
#   Needs:   Do you mean wallets that traded, or wallets that signed in?
edisyl ask growth-metrics "How many wallets that traded last week?" --wait

A progress line (phase, e.g. “drawing up the plan”) comes back alongside the status while work runs; read it, don’t branch on it. Its wording changes, and it’s absent whenever it would add nothing to status.

Reading one step in full

action plan gives you the board. To read a step’s full result (its brief, complete content, and artifacts), use action step <action-id> <step-id> with the step id from action plan. action events is the diagnostic layer underneath: every event the work emitted, unsummarized.

Cancellation

action cancel is a request. An Action already inside a step finishes that step first.

Scripting

task, teach, and ask all take --id-only (bare Action id, nothing else) and --wait with --timeout (default 5m) to block to terminal in-process. Every action and schedule read command takes --quiet, which drops hints and banners and keeps the data.