pi-herdsman 0.36.3 → 0.37.0
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/index.js +7741 -7241
- package/dist/index.js.map +4 -4
- package/dist/peer-bridge.js +3 -2
- package/dist/peer-bridge.js.map +2 -2
- package/docs/development/validation.md +8 -0
- package/docs/guides/project-orchestration.md +20 -0
- package/docs/reference/status-widget.md +55 -12
- package/docs/reference/supervisor.md +3 -0
- package/package.json +5 -2
|
@@ -118,6 +118,14 @@ npm run check
|
|
|
118
118
|
`npm run check` remains the bounded full check, including platform-appropriate
|
|
119
119
|
process-tree cleanup when the test runner times out or leaks descendants.
|
|
120
120
|
|
|
121
|
+
## Node compatibility target
|
|
122
|
+
|
|
123
|
+
`package.json` `engines.node` defines the minimum supported Node.js version.
|
|
124
|
+
Compatibility CI jobs and esbuild derive their targets from it; `@types/node`
|
|
125
|
+
must use the same major version. The Unix and Windows installers check the
|
|
126
|
+
published package requirement. Release/Preview tooling and the Docker image
|
|
127
|
+
may deliberately use newer Node versions.
|
|
128
|
+
|
|
121
129
|
## Dependency availability
|
|
122
130
|
|
|
123
131
|
A fresh worktree may not contain `node_modules`.
|
|
@@ -70,6 +70,20 @@ A branch is optional when starting new work. If omitted,
|
|
|
70
70
|
Pi Herdsman can generate one. The first delegation can create the first linked
|
|
71
71
|
worktree, and an unoccupied existing worktree for the branch is reused.
|
|
72
72
|
|
|
73
|
+
Manager's optional worktree statistics compare tracked working-tree contents
|
|
74
|
+
with the assignment's stable Git baseline, including committed, staged, and
|
|
75
|
+
unstaged changes without double-counting. Untracked files are excluded from
|
|
76
|
+
numeric totals but make the row dirty (`*`); binary-only changes can appear as
|
|
77
|
+
a changed-file count such as `2f`, and `±0` means no net tracked differences.
|
|
78
|
+
Baselines captured for new assignments stay fixed when Manager's `main`
|
|
79
|
+
advances, although merging upstream changes into a Lead checkout can contribute
|
|
80
|
+
to its net diff. Older assignments without a stored baseline use an approximate
|
|
81
|
+
merge-base estimate shown with `~`; it is not saved. If the Lead checks out a
|
|
82
|
+
different branch, `↪ branch` shows the actual branch while the original
|
|
83
|
+
assignment handle remains the row's focus target. These are advisory net
|
|
84
|
+
worktree measurements, not cumulative Lead-authored effort or proof of
|
|
85
|
+
integration readiness.
|
|
86
|
+
|
|
73
87
|
Start new work with `delegate_work`. Resume existing work with
|
|
74
88
|
`resume_work` using its original assignment branch handle; resume reuses the
|
|
75
89
|
existing verified checkout and exact Pi session without switching its current
|
|
@@ -211,6 +225,12 @@ reported as unresolved unless a later Lead execution completes successfully.
|
|
|
211
225
|
Cancellation is reported separately as a factual, nonterminal message without
|
|
212
226
|
partial assistant output.
|
|
213
227
|
|
|
228
|
+
Manager-bound Lead messages are labeled by source: explicit `message_supervisor`
|
|
229
|
+
messages, completed direct responses, settled herd runs, execution failures, and
|
|
230
|
+
cancellations have distinct headings. These labels describe the message only;
|
|
231
|
+
they do not imply assignment acceptance, review readiness, integration, or
|
|
232
|
+
retirement.
|
|
233
|
+
|
|
214
234
|
An assigned Lead reports an execution failure to its Manager through the
|
|
215
235
|
durable project-message path. If the Manager is absent, the report remains
|
|
216
236
|
pending; when the exact current Manager returns idle, an unreceived failure
|
|
@@ -63,8 +63,8 @@ a confidently identified mismatch.
|
|
|
63
63
|
|
|
64
64
|
```text
|
|
65
65
|
● manager · pi-herdsman
|
|
66
|
-
├─ ● feat/example
|
|
67
|
-
└─
|
|
66
|
+
├─ ● feat/example
|
|
67
|
+
└─ ◫ fix/other
|
|
68
68
|
```
|
|
69
69
|
|
|
70
70
|
When available, managed work rows can append the Lead's active herd-run elapsed
|
|
@@ -72,8 +72,8 @@ time and Pi's current context percentage:
|
|
|
72
72
|
|
|
73
73
|
```text
|
|
74
74
|
● manager · pi-herdsman
|
|
75
|
-
├─ ● feat/foo
|
|
76
|
-
└─
|
|
75
|
+
├─ ● feat/foo 14m 61%
|
|
76
|
+
└─ ◫ fix/other
|
|
77
77
|
```
|
|
78
78
|
|
|
79
79
|
Elapsed time is the same active herd-run interval shown by the ordinary Lead
|
|
@@ -84,20 +84,63 @@ the live Lead's runtime state. Chief rows receive matching theme styling but do
|
|
|
84
84
|
not show these telemetry columns. A stale Manager snapshot suppresses elapsed
|
|
85
85
|
telemetry while retaining other available metadata such as context percentage.
|
|
86
86
|
|
|
87
|
-
Manager work
|
|
88
|
-
|
|
89
|
-
|
|
90
|
-
|
|
91
|
-
|
|
92
|
-
verified build incompatibility
|
|
93
|
-
|
|
94
|
-
|
|
87
|
+
Routine Manager work is symbol-only: active work uses the verified Lead's
|
|
88
|
+
lifecycle marker, or muted `·` when no verified Lead is present; paused work
|
|
89
|
+
uses muted `◫`, distinct from an idle/done Lead's `○`. Exceptional states keep
|
|
90
|
+
their complete labels: blocked, unknown, and lost describe the verified Lead's
|
|
91
|
+
runtime state; project unknown and conflict use `?` and warning-colored `!`;
|
|
92
|
+
independently verified build incompatibility uses `!` and `build mismatch`.
|
|
93
|
+
These concise labels are presentation only and do not change work status or
|
|
94
|
+
focus eligibility. Ordinary supervised Lead rows retain `· Lead`, which
|
|
95
|
+
identifies their relationship rather than repeating lifecycle state.
|
|
96
|
+
|
|
97
|
+
Chief direct and nested Leads likewise use one marker before the identity;
|
|
98
|
+
only blocked, unknown, lost, and incompatible-build rows add explanatory text.
|
|
99
|
+
The ordinary Lead Agent widget retains a single animated/static lifecycle
|
|
100
|
+
marker. It adds aligned text only for blocked, unknown, and lost Agents, while
|
|
101
|
+
its ambient header summarizes nonzero working, blocked, transitioning,
|
|
102
|
+
unknown, and lost categories as `● N ◐ N ◌ N ? N × N` (starting and settling share
|
|
103
|
+
`◌`). If no Agent needs an attention label, the state-text column disappears;
|
|
104
|
+
when attention is present, routine rows keep its blank alignment. Running
|
|
105
|
+
menus and consequential confirmations continue to use full descriptive
|
|
106
|
+
lifecycle labels and counts.
|
|
95
107
|
Ordinary Leads supervised by the Manager are shown separately from project
|
|
96
108
|
work, with a supervised label; they are not treated as managed work. A Manager
|
|
97
109
|
can inspect an eligible direct report through the overview, independently of
|
|
98
110
|
mouse focus/navigation. Branches are display handles; exact session identity is
|
|
99
111
|
revalidated for control actions.
|
|
100
112
|
|
|
113
|
+
### Manager Git indicators
|
|
114
|
+
|
|
115
|
+
Eligible Manager work rows can show net tracked textual additions and
|
|
116
|
+
deletions from a stable assignment baseline to the current worktree contents.
|
|
117
|
+
The comparison includes committed, staged, and unstaged tracked changes without
|
|
118
|
+
double-counting them; untracked files do not contribute numeric totals. `*`
|
|
119
|
+
means the assigned checkout is dirty, including untracked changes. Thus `±0*`
|
|
120
|
+
means the verified comparison found no net tracked differences but the checkout
|
|
121
|
+
is dirty; `±0` means it found no net tracked differences and the checkout is
|
|
122
|
+
clean. A changed-file count such as `2f` is used when files changed without
|
|
123
|
+
textual line counts, such as binary-only changes. These are net worktree
|
|
124
|
+
contents, not cumulative effort or proof of merge readiness; merging upstream
|
|
125
|
+
changes into a Lead checkout can contribute to its net diff.
|
|
126
|
+
|
|
127
|
+
The stored baseline remains fixed when the Manager's `main` advances. Existing
|
|
128
|
+
assignments without a stored baseline use a dynamically inferred merge-base
|
|
129
|
+
approximation, marked with `~` before numeric statistics. That fallback is not
|
|
130
|
+
persisted. If a Lead switches branches in its assigned worktree, `↪` followed
|
|
131
|
+
by the checked-out branch distinguishes that mutable Git branch from the
|
|
132
|
+
original assignment branch shown as the row's clickable handle. Branch-switch
|
|
133
|
+
metadata and Git suffixes are not clickable; the assignment handle continues to
|
|
134
|
+
target the same exact managed Lead.
|
|
135
|
+
|
|
136
|
+
No Git suffix means the optional observation is unavailable or has not yet
|
|
137
|
+
been sampled; it does not mean the checkout is clean. Sampling is best-effort,
|
|
138
|
+
with a minimum interval of about 30 seconds between attempts. Under width
|
|
139
|
+
pressure, optional Git details may be omitted. The branch remains the
|
|
140
|
+
navigation target; its suffix is not clickable. Git indicators are advisory
|
|
141
|
+
display values only and do not establish project status, authority, or
|
|
142
|
+
integration readiness.
|
|
143
|
+
|
|
101
144
|
## Installation
|
|
102
145
|
|
|
103
146
|
The widget is installed only when the current session has a valid supported
|
|
@@ -37,6 +37,9 @@ clarification, blocker, warning, scope conflict, risk, decision request, or
|
|
|
37
37
|
important evidence. Continue independently safe work while a reply is pending.
|
|
38
38
|
Defer only steps that genuinely require the response; when no actionable work
|
|
39
39
|
remains, end the turn and rely on normal message delivery rather than polling.
|
|
40
|
+
For an assigned project Lead, this explicit message is labeled separately from
|
|
41
|
+
the automatic completed-result or settled-herd handoff; the label does not
|
|
42
|
+
change assignment lifecycle or review state.
|
|
40
43
|
|
|
41
44
|
For an assigned project Lead, information that can wait for the normal automatic
|
|
42
45
|
completed-result handoff should go there instead. Do not message solely to
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "pi-herdsman",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.37.0",
|
|
4
4
|
"type": "module",
|
|
5
5
|
"description": "Asynchronous Pi subagents and agent fleet orchestration for parallel coding agents with nested delegation, background work, and supervision in herdr.",
|
|
6
6
|
"license": "Apache-2.0",
|
|
@@ -69,6 +69,7 @@
|
|
|
69
69
|
"lint": "oxlint",
|
|
70
70
|
"check": "node scripts/check.mjs",
|
|
71
71
|
"build": "node scripts/build.mjs",
|
|
72
|
+
"typecheck": "tsc --project tsconfig.json",
|
|
72
73
|
"package:audit": "node scripts/package-audit.mjs",
|
|
73
74
|
"validate": "npm run format:check && npm run check && npm run build && npm run package:audit",
|
|
74
75
|
"release:check": "npm run validate && npm publish --dry-run --ignore-scripts",
|
|
@@ -86,11 +87,13 @@
|
|
|
86
87
|
"@earendil-works/pi-coding-agent": "1.4.0",
|
|
87
88
|
"@earendil-works/pi-tui": "1.4.0",
|
|
88
89
|
"@howaboua/pi-codex-conversion": "4.0.4",
|
|
90
|
+
"@types/node": "22.19.19",
|
|
89
91
|
"@types/proper-lockfile": "4.1.4",
|
|
90
92
|
"esbuild": "^0.28.2",
|
|
91
93
|
"oxlint": "1.90.0",
|
|
92
94
|
"prettier": "3.9.10",
|
|
93
|
-
"typebox": "^1.3.7"
|
|
95
|
+
"typebox": "^1.3.7",
|
|
96
|
+
"typescript": "7.0.2"
|
|
94
97
|
},
|
|
95
98
|
"engines": {
|
|
96
99
|
"node": ">=22.19.0"
|