niahere 0.2.64 → 0.2.65
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/package.json +1 -1
- package/src/core/consolidator.ts +15 -6
package/package.json
CHANGED
package/src/core/consolidator.ts
CHANGED
|
@@ -58,8 +58,9 @@ Nia uses a two-stage memory architecture. You are stage 1.
|
|
|
58
58
|
|
|
59
59
|
Your persona includes guidance to "save proactively" — that guidance applies
|
|
60
60
|
to LIVE chat, where you act on immediate user instruction. In THIS
|
|
61
|
-
consolidation pass,
|
|
62
|
-
|
|
61
|
+
consolidation pass, be selective but not paralyzed. If you see a genuine
|
|
62
|
+
learning, stage it. The promoter handles quality gating — your job is to
|
|
63
|
+
not miss real signals, not to be maximally conservative.
|
|
63
64
|
|
|
64
65
|
## Transcript
|
|
65
66
|
|
|
@@ -81,13 +82,20 @@ Answer these questions silently. If the answer to all of them is "nothing",
|
|
|
81
82
|
stop here and do not write anything.
|
|
82
83
|
|
|
83
84
|
1. What did the user correct, clarify, or teach you in this session?
|
|
85
|
+
(Includes: "no, do it this way", "don't use X", "always check Y first")
|
|
84
86
|
2. What NEW fact about the user, their projects, or their systems do you
|
|
85
87
|
now know that you did not at session start?
|
|
88
|
+
(Includes: architecture decisions, workflow patterns, tool preferences,
|
|
89
|
+
team structure, external system details discovered during task execution)
|
|
86
90
|
3. What decision was made that will constrain future work?
|
|
91
|
+
(Includes: "we're using X not Y", config changes, deployment patterns)
|
|
92
|
+
4. What did the user explicitly ask to be remembered?
|
|
87
93
|
|
|
88
|
-
Trivial small talk, greetings,
|
|
89
|
-
|
|
90
|
-
|
|
94
|
+
Trivial small talk, greetings, and pure status updates are NOT answers.
|
|
95
|
+
But corrections made DURING task execution ("no, check DynamoDB not S3"),
|
|
96
|
+
architecture learned while debugging ("ah, this service talks to X via Y"),
|
|
97
|
+
and workflow patterns revealed by how the user works — these ARE answers.
|
|
98
|
+
The bar is: would a fresh Nia session benefit from knowing this?
|
|
91
99
|
|
|
92
100
|
## Step 3 — Update staging.md
|
|
93
101
|
|
|
@@ -118,7 +126,8 @@ For each substantive answer:
|
|
|
118
126
|
- Do NOT write to \`memory.md\` or \`rules.md\`. Only the promoter job can.
|
|
119
127
|
- Do NOT use \`add_memory\` or \`add_rule\` MCP tools. Edit staging.md directly.
|
|
120
128
|
- Do NOT message the user.
|
|
121
|
-
-
|
|
129
|
+
- If nothing qualifies, do nothing. But don't be so conservative that the
|
|
130
|
+
pipeline starves — if you're skipping every session, your bar is too high.
|
|
122
131
|
|
|
123
132
|
Report a one-line summary of what you did: "staged N new / reinforced M /
|
|
124
133
|
skipped (trivial session)". No preamble.`;
|