@agimon-ai/doompi-workflow 0.0.1-alpha.0 → 0.0.1-alpha.2
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/README.md +1 -0
- package/dist/workflow/piExtension.cjs +8 -7
- package/dist/workflow/piExtension.cjs.map +1 -1
- package/dist/workflow/piExtension.mjs +8 -7
- package/dist/workflow/piExtension.mjs.map +1 -1
- package/dist/workflow/piTools.cjs +2 -2
- package/dist/workflow/piTools.cjs.map +1 -1
- package/dist/workflow/piTools.mjs +2 -2
- package/dist/workflow/piTools.mjs.map +1 -1
- package/dist/workflow/workflowInspector.cjs +1 -1
- package/dist/workflow/workflowInspector.mjs +1 -1
- package/dist/workflow/workflowRunPanel.cjs +1 -1
- package/dist/workflow/workflowRunPanel.mjs +1 -1
- package/package.json +5 -5
- package/skills/workflow-recovery/SKILL.md +15 -5
|
@@ -11,6 +11,11 @@ existing run from its active repair, preserving the worktree and the work alread
|
|
|
11
11
|
done. Choosing the wrong one either destroys progress or resumes a run that can
|
|
12
12
|
never succeed.
|
|
13
13
|
|
|
14
|
+
Recovery is also the explicit ownership-transfer path. A replacement Pi session
|
|
15
|
+
may recover a terminal failed run launched by an earlier session; the replay is
|
|
16
|
+
then stamped to the recovering session. This does not permit cross-session
|
|
17
|
+
control of running or completed work.
|
|
18
|
+
|
|
14
19
|
Read this before calling `workflow_run` with action `recover`.
|
|
15
20
|
|
|
16
21
|
## 1. Gather evidence first
|
|
@@ -23,8 +28,11 @@ Never act on the stage alone. Before touching a failed run, collect:
|
|
|
23
28
|
- **Only the repair referenced by `activeRepairId`.** Other repairs in the
|
|
24
29
|
directory belong to earlier cycles and are not yours to act on.
|
|
25
30
|
|
|
26
|
-
Use `workflow_run` with action `
|
|
27
|
-
|
|
31
|
+
Use `workflow_run` with action `recovery-evidence` for the terminal `run.json`
|
|
32
|
+
and durable `changelog.md`, `context.md`, and progress evidence. This dedicated
|
|
33
|
+
action is terminal-only and read-only, including for failures from an earlier
|
|
34
|
+
session; it never scrapes that session's live launcher. Generic `status` and
|
|
35
|
+
`tail` remain limited to runs owned by this session.
|
|
28
36
|
|
|
29
37
|
## 2. Choose exactly one outcome
|
|
30
38
|
|
|
@@ -62,12 +70,14 @@ Recovery and a fresh launch each consume one launch from the pass budget.
|
|
|
62
70
|
that is a fresh-launch case, not a reason to invent one.
|
|
63
71
|
- Prepare `run.json` **only** when verified recovery state requires it.
|
|
64
72
|
- Preserve the active repair and the existing worktree identity.
|
|
73
|
+
- Treat an “already claimed” result as another recovery winning the race; do
|
|
74
|
+
not launch fresh or retry around the claim.
|
|
65
75
|
- Do not use `--dry-run` for a real recovery.
|
|
66
76
|
|
|
67
77
|
## 5. Verify before reporting
|
|
68
78
|
|
|
69
79
|
Recover through the launcher recorded for the run. Then confirm **real process
|
|
70
80
|
and registry progress**. A `workflow_run` action `recover` call that returns is
|
|
71
|
-
not a recovery that succeeded.
|
|
72
|
-
|
|
73
|
-
anything still blocked.
|
|
81
|
+
not a recovery that succeeded. After ownership has transferred and the replacement run is visible to this
|
|
82
|
+
session, check `workflow_run` with action `status` before reporting an outcome.
|
|
83
|
+
Say what you recovered, what you skipped and why, and anything still blocked.
|