@sjawhar/pi-legion 18.50.3 → 18.51.0

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
@@ -14337,7 +14337,7 @@ import { logger as logger3 } from "@oh-my-pi/pi-utils";
14337
14337
  // package.json
14338
14338
  var package_default = {
14339
14339
  name: "@sjawhar/pi-legion",
14340
- version: "18.50.3",
14340
+ version: "18.51.0",
14341
14341
  license: "Apache-2.0",
14342
14342
  type: "module",
14343
14343
  omp: {
@@ -33,13 +33,16 @@ Launch, `gh`, pull requests, takeover: `skill://legion-controller/references/ses
33
33
 
34
34
  1. Recheck [trees waiting on a root claim](#trees-waiting-on-a-root-claim).
35
35
  2. Do the wake's work, or, on a start message, the start procedure below.
36
- 3. On your first turn of each UTC week, whatever woke you: run the weekly audit
36
+ 3. On your first turn of each six-hour UTC block, whatever woke you: run the consolidation pass
37
+ (`skill://legion-controller/references/consolidation.md`), because related issues left apart
38
+ wait for a person to ask for them to be folded.
39
+ 4. On your first turn of each UTC week, whatever woke you: run the weekly audit
37
40
  (`skill://legion-controller/references/weekly-audit.md`) before the day's report, so its
38
41
  numbers and its nudges go in that report.
39
- 4. On your first turn of each UTC day, whatever woke you: run the weekly-plan reconciliation
42
+ 5. On your first turn of each UTC day, whatever woke you: run the weekly-plan reconciliation
40
43
  (`skill://legion-controller/references/backlog-stewardship.md`), then post the
41
44
  [daily report](#daily-report).
42
- 5. Name the [free slots](#free-slots) in your summary, so a person sees room to hand Legion work.
45
+ 6. Name the [free slots](#free-slots) in your summary, so a person sees room to hand Legion work.
43
46
 
44
47
  ## What happened before you started
45
48
 
@@ -0,0 +1,115 @@
1
+ # The consolidation pass
2
+
3
+ Read this on your first turn of each six-hour UTC block, whatever woke you, and run both steps
4
+ below once the wake's work is done, at step 3 of the skill's every-turn steps. Finding issues that
5
+ are the same work, or that fail for one cause, and folding them is this pass's job, so that a
6
+ person never has to ask for it.
7
+
8
+ ## The block
9
+
10
+ The blocks start at 00:00, 06:00, 12:00 and 18:00 UTC: `date -u +%Y-%m-%dT%H:%M:%SZ` prints the
11
+ time now, and the current block started at the latest of those times at or before it. Read the
12
+ report issue (`skill://legion-controller/references/daily-report.md`), then its pass record:
13
+
14
+ ```bash
15
+ dispatch doc-read --issue <REPORT KEY> --artifact consolidation.md
16
+ ```
17
+
18
+ This block's pass has run when the record has a section headed `## Consolidation pass <block
19
+ start>`, whichever controller session wrote it, so a controller that took over from another does
20
+ not run the block again. A `not found` answer means no pass has run yet.
21
+
22
+ ## 1. Duplicates
23
+
24
+ For each open issue of the project, read the related items of its title and spec through the
25
+ no-write related route, leaving the issue, its ancestors and its descendants out, and judge
26
+ whether each pair the two issues' own records do not already settle is the same work.
27
+
28
+ The project holds hundreds of open issues and one related read takes about a second, so a serial
29
+ pass is ten minutes of HTTP before any judging. Shard the open list and dispatch one full agent per
30
+ shard of at most 50 issues, `task(agent="deep-worker")`, which can run commands and read Dispatch,
31
+ GitHub and the workspace, and whose model a pod configures (the `task` agent's it does not), then
32
+ `wait` for them: each returns the pairs it found with the records it read for each, and you check
33
+ each finding against those records before you fold it. A read-only `scout` cannot look up what a
34
+ finding rests on, so a shard a scout read does not count: dispatch it again to a full agent.
35
+
36
+ Give each agent the shard's keys and these commands:
37
+
38
+ ```bash
39
+ dispatch read --issue <KEY> # title, spec artifact, Links:, Referenced by:
40
+ dispatch doc-read --issue <KEY> # the spec text the related read uses
41
+ dispatch related --project <PROJECT> --title '<its title>' --text-file - \
42
+ --exclude-issue <KEY> --exclude-issue <each ancestor> --exclude-issue <each descendant>
43
+ ```
44
+
45
+ `--exclude-issue` is repeatable and takes up to 50 issues; the issue itself, its ancestors and its
46
+ descendants all go in one call. The read names both: a `child_of` row under `Links:` is the
47
+ issue's own parent, read again on that parent for the next ancestor up, and each `child_of` row
48
+ under `Referenced by:` is a child, read again on each child for the next level down. Ask each
49
+ agent for, per issue, every related item that is an issue with the reason it looks like the same
50
+ work and the records that show it, and nothing else: no comment, no close, no message.
51
+
52
+ A pair is settled, and is skipped rather than proposed, when the two issues already say so: one
53
+ was closed as a duplicate of the other, one was reopened after such a close, or one was made a
54
+ child of the other. Read the first two from `dispatch read --issue <KEY> --all-events`'s
55
+ `Events:`: the close's reason is the `message.created` line just before `issue.closed`, and a
56
+ reopen is an `issue.updated` line that moves the status off `done`. No `Events:` line reports a
57
+ parent change, so read the third from the same read's `Links:` and `Referenced by:`.
58
+
59
+ Fold each pair you judge the same work by the duplicate rule of `skill://dispatch-first`: keep the
60
+ issue with the spec, copy to it what only the other has, and close the other
61
+ (`--status done --reason 'Duplicate of dispatch://KEY.'`). An issue carrying the `legion` label,
62
+ or another holder's claim, is not yours to close: comment the kept issue's name on it instead. A
63
+ pair inside a Legion tree goes to that tree's architect with `envoy_publish`, which decides it.
64
+
65
+ The record (below) lists every pair you judged the same work, with what you did about it, and
66
+ every pair you skipped with what settles it: the event, or the parent link.
67
+
68
+ ## 2. Recurring causes
69
+
70
+ This step reads every issue changed in the last 24 hours, whatever its status, because a recently
71
+ closed issue or a failed attempt can show a cause too. Ask each agent of step 1 also about the
72
+ issues of its shard changed in that time, and put the issues closed in that time, which step 1
73
+ does not read, in more shards of at most 50:
74
+
75
+ ```bash
76
+ dispatch issues --project <PROJECT> --status done --updated-since "$(date -u -d '24 hours ago' +%Y-%m-%dT%H:%M:%SZ)" --all
77
+ ```
78
+
79
+ Ask each agent, for each of those issues that fails, what it fails on: the broken tool, check,
80
+ rule or component, the evidence in the issue that shows it (an error, a file, a command), and the
81
+ records it read. Ask for no groups, because issues that fail for one cause can fall in different
82
+ shards. Gather every shard's answers, check each against its records, and group the issues that
83
+ fail for one cause yourself; keep a group of three or more. A group that one fix clears entirely
84
+ is a set of duplicates: fold it as step 1 says. Keep for this step only a group whose issues each
85
+ need their own change beyond the shared cause.
86
+
87
+ Root each cause you keep once, whichever trees its issues sit in, as
88
+ `skill://legion-controller/references/finding-asks.md#hand-over-a-clear-defect` says. Post its
89
+ evidence comment on the report issue first, its first line `Clear defect from controller`, naming
90
+ the cause, each issue of the group and the shared evidence; catch-up step 4 of that reference
91
+ brings it back at a later start until it carries a reply. Then take its steps 2 to 5: comment on
92
+ the open issue that already tracks the cause, else hand it to the running tree whose code it is,
93
+ else file one issue for it, and reply under the evidence comment. The filed issue names each issue
94
+ of the group and the shared evidence, and carries the `Filed for` line of `finding-asks.md` for
95
+ the person who asked for this pass: `Filed for <them> at their request:
96
+ dispatch://INFRA-460/comment/67cf2b86-5c76-41a9-a53a-7bf2dce03602`. It takes the `legion` label
97
+ only when the fix needs no choice a person must make (that procedure's step 1); otherwise it waits
98
+ in triage for a person, and the reply is still `Filed dispatch://<KEY>`. Then comment on each
99
+ issue of the group the cause's issue, or the root of the tree it went to. Close no issue of the
100
+ group: each needs its own change as well.
101
+
102
+ ## The record
103
+
104
+ Record the pass in the report issue's `consolidation.md` artifact, not in a message: a message on
105
+ the report issue reaches a person's Inbox, and the day's report posts only while the issue holds
106
+ no message from that day. Add at the top of the record you read, or of a new one, a section headed
107
+ `## Consolidation pass <block start>` with three counts, each with its keys: the issues read; the
108
+ pairs folded (closed, commented or sent to an architect); and the causes rooted (issues commented,
109
+ handed or filed for a recurring cause). Then every pair and group you skipped, with what settles
110
+ it. Drop sections older than eight days, which the artifact's earlier versions keep, and upload
111
+ the whole record, which makes its next version:
112
+
113
+ ```bash
114
+ dispatch artifact --issue <REPORT KEY> --name consolidation.md --path <the record file>
115
+ ```
@@ -76,10 +76,11 @@ every issue by its `dispatch://` link and what it does for the reader, every pul
76
76
  - **The week** (on the report of your first turn of each UTC week only, from
77
77
  `skill://legion-controller/references/weekly-audit.md`): the window's two times; the forced
78
78
  creates by filer with the no-answer creates apart; the tree sizes in the five buckets with the
79
- parked descendants; each nudge you sent, by its tree; and the duplicate audit's pairs, those
80
- you folded and those you skipped with the event that settles each. Name beside each number the
81
- command that recounts it. When you lead a stream that keeps a scoreboard, the same two measures
82
- go on it under `forced_creates_by_filer` and `tree_sizes`.
79
+ parked descendants; each nudge you sent, by its tree; and the consolidation passes' three counts
80
+ summed over the week, with the link of the report issue's `consolidation.md` artifact, which
81
+ lists each pair and group with what was done about it or what settles it. Name beside each
82
+ number the command that recounts it. When you lead a stream that keeps a scoreboard, the same
83
+ two measures go on it under `forced_creates_by_filer` and `tree_sizes`.
83
84
 
84
85
  A day with nothing finished says so in one sentence. When the lists do not fit, give each list's
85
86
  count and name its issues of `--priority 0` and `--priority 1`, in board rank order.
@@ -1,8 +1,8 @@
1
1
  # The weekly audit
2
2
 
3
- Read this on your first turn of each UTC week, whatever woke you, and run all four passes below
4
- before you post the day's report. They are the week's own numbers, beside the daily ones. The
5
- fourth pass, the duplicate audit, serves R8 of dispatch://INFRA-1160 and is not optional.
3
+ Read this on your first turn of each UTC week, whatever woke you, and run the three passes below,
4
+ and sum the fourth, before you post the day's report. They are the week's own numbers, beside the
5
+ daily ones.
6
6
 
7
7
  ## The week
8
8
 
@@ -104,43 +104,7 @@ List every nudge you sent in the day's report.
104
104
 
105
105
  ## 4. The duplicate audit
106
106
 
107
- For each open issue of the project, read the related items of its title and spec through the
108
- no-write related route, leaving the issue, its ancestors and its descendants out, and judge
109
- whether each pair the two issues' own records do not already settle is the same work.
110
-
111
- The project holds hundreds of open issues and one related read takes about a second, so a serial
112
- pass is ten minutes of HTTP before any judging. Shard the open list and dispatch one
113
- `task(agent="scout")` per shard of at most 50 issues, then `wait` for them: each returns the pairs
114
- it found with its evidence, and you judge and fold them yourself.
115
-
116
- Give each scout the shard's keys and these commands:
117
-
118
- ```bash
119
- dispatch read --issue <KEY> # title, spec artifact, Links:, Referenced by:
120
- dispatch doc-read --issue <KEY> # the spec text the related read uses
121
- dispatch related --project <PROJECT> --title '<its title>' --text-file - \
122
- --exclude-issue <KEY> --exclude-issue <each ancestor> --exclude-issue <each descendant>
123
- ```
124
-
125
- `--exclude-issue` is repeatable and takes up to 50 issues; the issue itself, its ancestors and its
126
- descendants all go in one call. The read names both: a `child_of` row under `Links:` is the
127
- issue's own parent, read again on that parent for the next ancestor up, and each `child_of` row
128
- under `Referenced by:` is a child, read again on each child for the next level down. Ask each
129
- scout for, per issue, every related item that is an issue with the reason it looks like the same
130
- work, and nothing else: no comment, no close, no message.
131
-
132
- A pair is settled, and is skipped rather than proposed, when the two issues already say so: one
133
- was closed as a duplicate of the other, one was reopened after such a close, or one was made a
134
- child of the other. Read the first two from `dispatch read --issue <KEY> --all-events`'s
135
- `Events:`: the close's reason is the `message.created` line just before `issue.closed`, and a
136
- reopen is an `issue.updated` line that moves the status off `done`. No `Events:` line reports a
137
- parent change, so read the third from the same read's `Links:` and `Referenced by:`.
138
-
139
- Fold each pair you judge the same work by the duplicate rule of `skill://dispatch-first`: keep the
140
- issue with the spec, copy to it what only the other has, and close the other
141
- (`--status done --reason 'Duplicate of dispatch://KEY.'`). An issue carrying the `legion` label,
142
- or another holder's claim, is not yours to close: comment the kept issue's name on it instead. A
143
- pair inside a Legion tree goes to that tree's architect with `envoy_publish`, which decides it.
144
-
145
- The report lists every pair you judged the same work, with what you did about it, and every pair
146
- you skipped with what settles it: the event, or the parent link.
107
+ The consolidation pass runs it every six hours
108
+ (`skill://legion-controller/references/consolidation.md`). The week's report sums the three counts
109
+ over the week's sections of the report issue's `consolidation.md` artifact (`dispatch doc-read
110
+ --issue <REPORT KEY> --artifact consolidation.md`). It serves R8 of dispatch://INFRA-1160.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@sjawhar/pi-legion",
3
- "version": "18.50.3",
3
+ "version": "18.51.0",
4
4
  "license": "Apache-2.0",
5
5
  "type": "module",
6
6
  "omp": {