@crewx/dreaming 0.1.0-rc.63 → 0.1.0-rc.65
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/SKILL.md +72 -71
- package/package.json +3 -3
package/SKILL.md
CHANGED
|
@@ -1,142 +1,143 @@
|
|
|
1
1
|
---
|
|
2
2
|
name: crewx/dreaming
|
|
3
3
|
description: >
|
|
4
|
-
|
|
5
|
-
|
|
4
|
+
Essential retrospective skill invoked selectively. A first-class self-reflection
|
|
5
|
+
tool for reviewing agent behavior at the task/conversation level. On par with
|
|
6
|
+
memory/doc/search — the agent judges the situation and calls it proactively.
|
|
6
7
|
metadata:
|
|
7
8
|
version: 0.1.0
|
|
8
9
|
built-in: true
|
|
9
10
|
---
|
|
10
11
|
|
|
11
|
-
# Dreaming —
|
|
12
|
+
# Dreaming — Essential Retrospective Skill
|
|
12
13
|
|
|
13
|
-
**dreaming
|
|
14
|
-
memory(
|
|
15
|
-
|
|
14
|
+
**dreaming is the agent's first-class self-reflection tool.**
|
|
15
|
+
On par with memory (store/search memories), doc (explore design docs), and search (search code),
|
|
16
|
+
the agent judges the situation and **calls it on its own** — it doesn't wait for a manual request.
|
|
16
17
|
|
|
17
|
-
##
|
|
18
|
+
## Two Views
|
|
18
19
|
|
|
19
|
-
|
|
|
20
|
+
| Command | View | When to use |
|
|
20
21
|
|--------|------|---------|
|
|
21
|
-
| `crewx dreaming --task=tsk_xxx` |
|
|
22
|
-
| `crewx dreaming --thread=thd_xxx --recent=N` |
|
|
23
|
-
| `crewx dreaming --thread=thd_xxx --agent=self --continue` |
|
|
22
|
+
| `crewx dreaming --task=tsk_xxx` | Magnifier | Deep dive into one task — action timeline, struggle patterns, efficiency patterns, skill usage |
|
|
23
|
+
| `crewx dreaming --thread=thd_xxx --recent=N` | Map | List of the last N tasks in the conversation + each one's end status |
|
|
24
|
+
| `crewx dreaming --thread=thd_xxx --agent=self --continue` | Continue | Detail on my most recent task + rule-based summary of the N before it |
|
|
24
25
|
|
|
25
|
-
##
|
|
26
|
+
## When To Call It (Agent Judgment Criteria)
|
|
26
27
|
|
|
27
|
-
-
|
|
28
|
-
-
|
|
29
|
-
-
|
|
30
|
-
-
|
|
28
|
+
- **Before starting work**: Check whether a task was just interrupted → check the `--thread` list
|
|
29
|
+
- **Retrying a failed task**: See where it stopped last time → check the timeline with `--task`
|
|
30
|
+
- **Understanding conversation context**: Get a list of what was asked in this conversation → `--thread --recent=5`
|
|
31
|
+
- **Self-check after a task**: Check for struggle patterns / efficiency patterns → retrospect with `--task`
|
|
31
32
|
|
|
32
|
-
##
|
|
33
|
+
## Usage
|
|
33
34
|
|
|
34
35
|
```bash
|
|
35
|
-
#
|
|
36
|
+
# Deep retrospective on one task
|
|
36
37
|
crewx dreaming --task=tsk_abc123
|
|
37
38
|
|
|
38
|
-
#
|
|
39
|
+
# List of the last 5 in the conversation (default)
|
|
39
40
|
crewx dreaming --thread=thd_xyz789
|
|
40
41
|
|
|
41
|
-
#
|
|
42
|
+
# See more
|
|
42
43
|
crewx dreaming --thread=thd_xyz789 --recent=20
|
|
43
44
|
|
|
44
|
-
#
|
|
45
|
+
# Drill down into a specific task
|
|
45
46
|
crewx dreaming --task=tsk_abc123 --json
|
|
46
47
|
```
|
|
47
48
|
|
|
48
|
-
##
|
|
49
|
+
## Output Example — --task Mode
|
|
49
50
|
|
|
50
51
|
```
|
|
51
52
|
━━━ Dreaming: task tsk_abc123 ━━━
|
|
52
53
|
Agent: core_dev (acp/claude) | Status: success | Duration: 312s | Tools: 47 calls
|
|
53
|
-
📋
|
|
54
|
-
📝
|
|
54
|
+
📋 Instruction: promote dreaming to a built-in skill
|
|
55
|
+
📝 Conclusion: packages/built-in/dreaming/ created successfully
|
|
55
56
|
|
|
56
|
-
[
|
|
57
|
+
[Timeline]
|
|
57
58
|
1. Read packages/built-in/memory/package.json
|
|
58
59
|
2. Read packages/built-in/doc/tsconfig.json
|
|
59
60
|
3. Edit packages/built-in/dreaming/package.json
|
|
60
61
|
...
|
|
61
62
|
41. Bash npx tsc --noEmit → ❌ error TS2345
|
|
62
|
-
42. Edit packages/built-in/dreaming/src/analyzer.ts ←
|
|
63
|
+
42. Edit packages/built-in/dreaming/src/analyzer.ts ← re-edited the same file
|
|
63
64
|
43. Bash npx tsc --noEmit → ✅ pass
|
|
64
|
-
💡
|
|
65
|
+
💡 Struggle: edited analyzer.ts twice (41→43) — type mismatch
|
|
65
66
|
|
|
66
|
-
[
|
|
67
|
-
crewx/memory
|
|
68
|
-
crewx/doc
|
|
67
|
+
[Skill Usage]
|
|
68
|
+
crewx/memory 3x (index, find, save)
|
|
69
|
+
crewx/doc 2x (toc, section)
|
|
69
70
|
|
|
70
|
-
[
|
|
71
|
-
⚠️
|
|
72
|
-
💡
|
|
73
|
-
🔍
|
|
71
|
+
[Retrospective Summary]
|
|
72
|
+
⚠️ Struggle pattern: repeated edits to the same file, 1 case (analyzer.ts x2)
|
|
73
|
+
💡 Insight: most errors were type mismatches (1 case)
|
|
74
|
+
🔍 Used code analysis — tsserver called 5x
|
|
74
75
|
```
|
|
75
76
|
|
|
76
|
-
##
|
|
77
|
+
## Output Example — --thread Mode
|
|
77
78
|
|
|
78
79
|
```
|
|
79
|
-
━━━ Dreaming: thread thd_xyz789 —
|
|
80
|
-
|
|
80
|
+
━━━ Dreaming: thread thd_xyz789 — last 5 ━━━
|
|
81
|
+
Task list: 5
|
|
81
82
|
|
|
82
|
-
1. ✅ tsk_abc123 | core_dev | success | 312s | dreaming
|
|
83
|
-
2. ❌ tsk_def456 | core_dev | failed | 45s | tsc
|
|
84
|
-
3. ✅ tsk_ghi789 | ui_lead | completed | 180s | UI
|
|
85
|
-
4. ✅ tsk_jkl012 | core_dev | success | 240s | SDK
|
|
86
|
-
5. ✅ tsk_mno345 | core_dev | success | 90s | CLI registry
|
|
83
|
+
1. ✅ tsk_abc123 | core_dev | success | 312s | promote dreaming to a built-in skill
|
|
84
|
+
2. ❌ tsk_def456 | core_dev | failed | 45s | tsc build verification
|
|
85
|
+
3. ✅ tsk_ghi789 | ui_lead | completed | 180s | UI component review
|
|
86
|
+
4. ✅ tsk_jkl012 | core_dev | success | 240s | apply SDK repository pattern
|
|
87
|
+
5. ✅ tsk_mno345 | core_dev | success | 90s | register CLI registry
|
|
87
88
|
|
|
88
|
-
|
|
89
|
+
Drill down: crewx dreaming --task=<id>
|
|
89
90
|
```
|
|
90
91
|
|
|
91
|
-
##
|
|
92
|
+
## Output Example — --continue Mode (Continue My Previous Work)
|
|
92
93
|
|
|
93
|
-
|
|
94
|
+
Within the same thread, this picks out only the **current agent's own** previous work, showing the most recent one in detail plus rule-based one-line summaries of the ones before it. The summaries are based on tool count / status / duration / files, not an LLM.
|
|
94
95
|
|
|
95
96
|
```
|
|
96
|
-
━━━
|
|
97
|
+
━━━ Continue My Previous Work ━━━
|
|
97
98
|
Thread: thd_xyz789 | Agent: core_lead
|
|
98
99
|
|
|
99
|
-
[
|
|
100
|
+
[Most Recent Task]
|
|
100
101
|
|
|
101
102
|
━━━ Dreaming: task tsk_123 ━━━
|
|
102
|
-
...
|
|
103
|
+
... detailed activity log (timeline capped at 30, remove with --full) ...
|
|
103
104
|
|
|
104
|
-
[
|
|
105
|
-
- tsk_122 | ✅ success | 120s | Bash
|
|
106
|
-
- tsk_121 | ❌ failed | 45s | Bash
|
|
105
|
+
[Previous Task Summary]
|
|
106
|
+
- tsk_122 | ✅ success | 120s | Bash 3x, Read 4x, Edit 0x
|
|
107
|
+
- tsk_121 | ❌ failed | 45s | Bash 2x | ⚠️ struggle
|
|
107
108
|
|
|
108
|
-
━━━
|
|
109
|
-
|
|
109
|
+
━━━ Next Actions ━━━
|
|
110
|
+
Full activity log:
|
|
110
111
|
crewx dreaming --task=tsk_122 --full
|
|
111
|
-
|
|
112
|
+
Detailed task log JSON:
|
|
112
113
|
crewx dreaming --task=tsk_122 --json
|
|
113
|
-
|
|
114
|
+
Task list for this conversation:
|
|
114
115
|
crewx dreaming --thread=thd_xyz789 --agent=core_lead
|
|
115
116
|
```
|
|
116
117
|
|
|
117
118
|
```bash
|
|
118
|
-
#
|
|
119
|
+
# Continue the current agent's own previous work
|
|
119
120
|
crewx dreaming --thread=thd_xyz789 --agent=self --continue
|
|
120
121
|
|
|
121
|
-
#
|
|
122
|
+
# Explicit agent + last 3 (most recent 1 detailed + previous 2 summarized)
|
|
122
123
|
crewx dreaming --thread=thd_xyz789 --agent=core_lead --continue --recent=3
|
|
123
124
|
```
|
|
124
125
|
|
|
125
|
-
> `--agent=self`
|
|
126
|
+
> `--agent=self` resolves the current agent via the `CREWX_AGENT_ID` environment variable. If the variable isn't set, you must specify an explicit agent id.
|
|
126
127
|
|
|
127
128
|
## Options
|
|
128
129
|
|
|
129
|
-
|
|
|
130
|
+
| Option | Default | Description |
|
|
130
131
|
|------|--------|------|
|
|
131
|
-
| `--task=ID` | — |
|
|
132
|
-
| `--thread=ID` | — |
|
|
133
|
-
| `--agent=ID` | — |
|
|
134
|
-
| `--continue` | false |
|
|
135
|
-
| `--recent=N` |
|
|
136
|
-
| `--full` | false |
|
|
137
|
-
| `--json` | false |
|
|
138
|
-
|
|
139
|
-
## CAL "
|
|
140
|
-
|
|
141
|
-
|
|
142
|
-
|
|
132
|
+
| `--task=ID` | — | Task ID (tsk_xxx) |
|
|
133
|
+
| `--thread=ID` | — | Thread ID (thd_xxx) |
|
|
134
|
+
| `--agent=ID` | — | Only this agent within the thread (`self` → current agent) |
|
|
135
|
+
| `--continue` | false | Continue my previous work (most recent 1 detailed + previous N rule-based summary) |
|
|
136
|
+
| `--recent=N` | list 5 / continue 3 | Last N items |
|
|
137
|
+
| `--full` | false | Remove the timeline cap + include raw fields in JSON |
|
|
138
|
+
| `--json` | false | Pure JSON output (no CAL footer) |
|
|
139
|
+
|
|
140
|
+
## CAL "Next Actions"
|
|
141
|
+
|
|
142
|
+
A CAL footer guiding the next action is appended to the bottom of human-readable output (`--full`, `--json`, thread drill-down).
|
|
143
|
+
**The `--json` output has no footer and can be parsed as a single JSON value.** Without `--full`, the raw input/result fields are excluded from the JSON.
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@crewx/dreaming",
|
|
3
|
-
"version": "0.1.0-rc.
|
|
3
|
+
"version": "0.1.0-rc.65",
|
|
4
4
|
"description": "Dreaming — mandatory retrospection skill for CrewX agents (task/thread review)",
|
|
5
5
|
"type": "commonjs",
|
|
6
6
|
"main": "dist/cli.js",
|
|
@@ -31,8 +31,8 @@
|
|
|
31
31
|
],
|
|
32
32
|
"license": "MIT",
|
|
33
33
|
"dependencies": {
|
|
34
|
-
"@crewx/
|
|
35
|
-
"@crewx/
|
|
34
|
+
"@crewx/shared": "0.0.6",
|
|
35
|
+
"@crewx/sdk": "0.9.0-rc.40"
|
|
36
36
|
},
|
|
37
37
|
"devDependencies": {
|
|
38
38
|
"@types/node": "^20.0.0",
|