@sjawhar/pi-legion 19.13.0 → 19.13.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/legion.js
CHANGED
|
@@ -28516,7 +28516,7 @@ import { logger as logger3 } from "@oh-my-pi/pi-utils";
|
|
|
28516
28516
|
// package.json
|
|
28517
28517
|
var package_default = {
|
|
28518
28518
|
name: "@sjawhar/pi-legion",
|
|
28519
|
-
version: "19.13.
|
|
28519
|
+
version: "19.13.1",
|
|
28520
28520
|
license: "Apache-2.0",
|
|
28521
28521
|
type: "module",
|
|
28522
28522
|
omp: {
|
|
@@ -28528,7 +28528,7 @@ var package_default = {
|
|
|
28528
28528
|
]
|
|
28529
28529
|
},
|
|
28530
28530
|
legion: {
|
|
28531
|
-
daemonApiVersion:
|
|
28531
|
+
daemonApiVersion: 104
|
|
28532
28532
|
},
|
|
28533
28533
|
repository: {
|
|
28534
28534
|
type: "git",
|
|
@@ -30442,7 +30442,7 @@ function createLegionTool(deps) {
|
|
|
30442
30442
|
description: legionToolDescription({
|
|
30443
30443
|
envoyPublish: "envoy_publish",
|
|
30444
30444
|
before: PHASE_OP_SENTENCES.readRecord + PHASE_OP_SENTENCES.readReleaseRecords,
|
|
30445
|
-
after: PHASE_OP_SENTENCES.requestBackwardMove + "retry_or_escalate decides an issue the daemon holds after its worker failed (issue, decision): retry puts the issue " + "back in the phase it was held from and starts that phase's worker again, escalate leaves it held for the controller, " + "and an optional reason says why you decided as you did: a retry writes it into the restarted worker's task and the " + "daemon's own log line for the release, and an escalation sends it to the controller with the hold. " + "sign_off is the architect's acceptance of an issue in accepting, moving it to retro; the daemon refuses it in every " + "other phase (SIGNOFF_OUTSIDE_ACCEPTING) and while an issue beneath it is still in the workflow " + "(SIGNOFF_CHILD_IN_WORKFLOW, naming each one), since a parent's head proves its children's merged work: run each one " + "to done, and sign off again. " + "Park one with park_child only when a person asked for that move, or when the work is done, folded into another issue or no longer needed, and say in a comment which of those it is. " + "Ask the person about a child that can reach neither done nor a park under that rule, and hold the acceptance until they answer. " + "close_root is refused the same way and with the same code while an issue beneath the root is still in the workflow, " + "since closing it would stop that work mid-turn: answer it the same way, park_child under the same rule included. " + "release_children and rerun_child are refused for a child whose ancestor is in retro or merging " + "(CHILD_UNDER_ACCEPTED_PARENT, naming the ancestor), since that ancestor's approved head cannot carry work released " + "under it now: wait for the ancestor to close, then release it. " + "The controller's read_state with no field returns a summary of the daemon's whole state \u2014 the project, admission, each " + "capability, one line per recorded issue, and under each issue Legion still runs its slot, hold, gate and each claim's " + "state with its last turn times \u2014 and with issue, issues, phase or settings returns the full records it names; " + "set_status moves an issue to todo, backlog or icebox. " + "Move an issue to backlog or icebox only when a person asked for that move, or when the work is done, folded into another issue or no longer needed, and cite that reason in a comment on the issue. " + "Two writes are not that move: a root's restart, which sets backlog and then todo in the same turn, and the " + "controller's report issue's icebox, because that issue holds no work. " + "fresh_session (issue, role) has one claim leave the session it runs in behind and start again in a new one: the daemon " + "stops the agent if it runs, forgets the session, and starts the claim, which loses everything that session held. A " + "phase worker that comes back owing no task is started through its issue's current phase, so it has that phase's work. " + "The answer is empty when the claim came back and nothing else is to be said
|
|
30445
|
+
after: PHASE_OP_SENTENCES.requestBackwardMove + "retry_or_escalate decides an issue the daemon holds after its worker failed (issue, decision): retry puts the issue " + "back in the phase it was held from and starts that phase's worker again, escalate leaves it held for the controller, " + "and an optional reason says why you decided as you did: a retry writes it into the restarted worker's task and the " + "daemon's own log line for the release, and an escalation sends it to the controller with the hold. " + "sign_off is the architect's acceptance of an issue in accepting, moving it to retro; the daemon refuses it in every " + "other phase (SIGNOFF_OUTSIDE_ACCEPTING) and while an issue beneath it is still in the workflow " + "(SIGNOFF_CHILD_IN_WORKFLOW, naming each one), since a parent's head proves its children's merged work: run each one " + "to done, and sign off again. " + "Park one with park_child only when a person asked for that move, or when the work is done, folded into another issue or no longer needed, and say in a comment which of those it is. " + "Ask the person about a child that can reach neither done nor a park under that rule, and hold the acceptance until they answer. " + "close_root is refused the same way and with the same code while an issue beneath the root is still in the workflow, " + "since closing it would stop that work mid-turn: answer it the same way, park_child under the same rule included. " + "release_children and rerun_child are refused for a child whose ancestor is in retro or merging " + "(CHILD_UNDER_ACCEPTED_PARENT, naming the ancestor), since that ancestor's approved head cannot carry work released " + "under it now: wait for the ancestor to close, then release it. " + "The controller's read_state with no field returns a summary of the daemon's whole state \u2014 the project, admission, each " + "capability, one line per recorded issue, and under each issue Legion still runs its slot, hold, gate and each claim's " + "state with its last turn times \u2014 and with issue, issues, phase or settings returns the full records it names; " + "set_status moves an issue to todo, backlog or icebox. " + "Move an issue to backlog or icebox only when a person asked for that move, or when the work is done, folded into another issue or no longer needed, and cite that reason in a comment on the issue. " + "Two writes are not that move: a root's restart, which sets backlog and then todo in the same turn, and the " + "controller's report issue's icebox, because that issue holds no work. " + "fresh_session (issue, role) has one claim leave the session it runs in behind and start again in a new one: the daemon " + "stops the agent if it runs, forgets the session, and starts the claim, which loses everything that session held. A " + "phase worker that comes back owing no task is started through its issue's current phase, so it has that phase's work. " + "The answer is empty when the claim came back and nothing else is to be said. phaseNotRestarted says why nothing " + "restarted the claim's phase, where the claim came back owing no task and the workflow refused that restart: the " + "claim is running, and no work was handed to it. It says what was refused, where this request stopped the claim " + "and a step that would have started it again was refused, and the state that left the claim in: read that state " + "in read_state, and ask for a fresh session again only where the claim is suspended. " + "It reaches a claim stored working that no process stands behind: the daemon reads that claim's process when the " + "request arrives, suspends the claim where it stands, and starts it again with the task its dead turn held. " + "The daemon refuses it for a claim whose process that read finds running " + "(CLAIM_IN_A_TURN), for a claim whose session is being compacted, which is not read at " + "all (CLAIM_IN_A_TURN), for a claim whose process the runtime would not answer for, for " + "one that records no process to read, and for one that moved under the request and that a second read " + "finds no longer working or gone once more, each left exactly as the request found it " + "(CLAIM_UNSETTLED; a claim that moved is otherwise answered as that second read finds it), for a claim a " + "suspension is held over, in whatever state the hold found it (CLAIM_IN_A_TURN where " + "the claim is working, CLAIM_UNSETTLED where it is not), for a claim between states \u2014 " + "one queued, one whose launch is in flight, and one left working under a relaunch the " + "wake pacing parked (CLAIM_UNSETTLED) \u2014 and for role controller. " + "compact (issue, role) shortens one claim's session instead: the daemon has Oh My Pi write a text summary of the " + "session so far, and the claim keeps its session, its workspace and everything it knows of its issue and is sent " + "whatever task it owes again. Try compact first for a claim whose session a provider keeps refusing, and " + "fresh_session for a claim no process stands behind, which compact refuses and fresh_session reaches. The answer " + "says whether the session was compacted and, when it was not, why; the claim is back either way. The daemon " + "refuses it for a claim that is suspended, failed or retired, and for one stored working " + "that no process stands behind (CLAIM_NOT_RUNNING: fresh_session reaches both), for a " + "claim whose process the daemon reads as running (CLAIM_IN_A_TURN), for a claim whose " + "session is being compacted already (CLAIM_IN_A_TURN), for a claim whose process the " + "runtime would not answer for, for one that records no process to read, and for one " + "that moved out of working while the request was made " + "(CLAIM_UNSETTLED), for a claim a " + "suspension is held over (CLAIM_IN_A_TURN where the claim is working, CLAIM_UNSETTLED " + "where it is not), for a claim between states \u2014 one queued, one whose launch is in " + "flight, one whose agent has registered and is not ready yet, which fresh_session " + "reaches, and one left working under a relaunch the wake pacing parked " + "(CLAIM_UNSETTLED) \u2014 for a claim whose role runs on Claude Code (CLAIM_NOT_ON_OMP) " + "and for role controller. " + "Each refusal says what the claim does now and which operation reaches each state. It " + "names no moment to come back at: ask again later, and read the claim's state in " + "read_state. "
|
|
30446
30446
|
}),
|
|
30447
30447
|
defaultInactive: true,
|
|
30448
30448
|
parameters: toolSchema(pi),
|
|
@@ -400,16 +400,31 @@ chain refuses the same work. Either way, `envoy_publish` the key, the role and t
|
|
|
400
400
|
issue's architect, which is who can narrow the work. Re-admit nothing and park nothing, because
|
|
401
401
|
the task is not lost and the session still runs.
|
|
402
402
|
|
|
403
|
-
One claim that stays stuck after that is yours to act on, and
|
|
404
|
-
|
|
403
|
+
One claim that stays stuck after that is yours to act on, and which door you use depends on what
|
|
404
|
+
the claim is doing. `compact` is the first thing for a claim whose session the provider keeps
|
|
405
|
+
refusing: it has the daemon shorten that claim's Oh My Pi session to a text summary and send it
|
|
405
406
|
whatever task it owed again, and the claim keeps its session, its workspace and everything it
|
|
406
407
|
knows of its issue. The answer says whether the session was compacted and, when it was not, why;
|
|
407
|
-
the claim is back either way, and whatever task it owed has gone again. Use `fresh_session`
|
|
408
|
-
|
|
408
|
+
the claim is back either way, and whatever task it owed has gone again. Use `fresh_session` once
|
|
409
|
+
the claim is still refused after a compaction, because a new session loses everything the old
|
|
409
410
|
one held — what the worker learned of its issue, and what it had not yet pushed or written down —
|
|
410
411
|
and cures nothing in a provider outage, where the next session's turns are refused as the last
|
|
411
412
|
one's were.
|
|
412
413
|
|
|
414
|
+
`compact` is the wrong door for a claim no process stands behind, which is a role that died: it
|
|
415
|
+
refuses such a claim `CLAIM_NOT_RUNNING` and names `fresh_session`, because it has no agent to
|
|
416
|
+
ask. Ask for `fresh_session` there. It is the one way back for a dead role, a dead architect of
|
|
417
|
+
your project included: it suspends the claim where your request arrives, leaves the session that
|
|
418
|
+
claim recorded behind, and starts the claim again with the task its dead turn held. Both doors
|
|
419
|
+
read the claim's process at the moment you ask, not the daemon's memory of an older report,
|
|
420
|
+
which a daemon restart forgets. A refusal of `CLAIM_IN_A_TURN` is a claim whose process that
|
|
421
|
+
read found running, or one whose session is being compacted. A claim whose process the runtime
|
|
422
|
+
would not answer for, one that records no process to read, and one that moved under your
|
|
423
|
+
request and that a second read finds no longer working or gone once more are refused
|
|
424
|
+
`CLAIM_UNSETTLED`, each left exactly as your request found it, so ask again once the runtime
|
|
425
|
+
answers; a claim that moved is otherwise answered as that second read finds it. A claim under a relaunch the wake pacing parks is refused
|
|
426
|
+
`CLAIM_UNSETTLED` too.
|
|
427
|
+
|
|
413
428
|
## GitHub mentions
|
|
414
429
|
|
|
415
430
|
Read the mention and its artifact. Answer a mention that asks you for triage or for information
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@sjawhar/pi-legion",
|
|
3
|
-
"version": "19.13.
|
|
3
|
+
"version": "19.13.1",
|
|
4
4
|
"license": "Apache-2.0",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"omp": {
|
|
@@ -12,7 +12,7 @@
|
|
|
12
12
|
]
|
|
13
13
|
},
|
|
14
14
|
"legion": {
|
|
15
|
-
"daemonApiVersion":
|
|
15
|
+
"daemonApiVersion": 104
|
|
16
16
|
},
|
|
17
17
|
"repository": {
|
|
18
18
|
"type": "git",
|