@aman_asmuei/aman-agent 0.25.0 → 0.26.0

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.
Files changed (4) hide show
  1. package/README.md +17 -13
  2. package/dist/index.js +648 -175
  3. package/dist/index.js.map +1 -1
  4. package/package.json +1 -1
package/README.md CHANGED
@@ -47,7 +47,7 @@
47
47
  <details>
48
48
  <summary><strong>Table of Contents</strong></summary>
49
49
 
50
- - [What's New](#whats-new-in-v0240)
50
+ - [What's New](#whats-new-in-v0260)
51
51
  - [The Problem](#the-problem)
52
52
  - [The Solution](#the-solution)
53
53
  - [Quick Start](#quick-start)
@@ -80,26 +80,30 @@
80
80
 
81
81
  ---
82
82
 
83
- ## What's New in v0.24.0
83
+ ## What's New in v0.26.0
84
84
 
85
- > **Session telemetry, post-mortem analysis, and a smarter feedback loop.**
85
+ > **Your agent learns new skills from its own sessions.**
86
86
 
87
- The agent now passively observes what happens during a session — tool calls, file changes, sentiment shifts, blockers, milestones — and on session end can generate a structured post-mortem with actionable patterns stored back as memories.
87
+ When a post-mortem runs, the LLM now proposes 0–2 reusable procedures it noticed you demonstrated. You accept or reject each one via a simple prompt at session end. Accepted skills are written to `~/.askill/skills.md` with embedded trigger keywords and **auto-activate** on subsequent sessions — zero extra LLM calls.
88
88
 
89
89
  | Feature | What it does |
90
90
  |:---|:---|
91
- | **Observation system** | Passive event capture: tool calls, errors, file changes, sentiment, topic shifts, blockers, milestones — all written to JSONL, never blocking the user |
92
- | **Post-mortem reports** | LLM-generated session analysis with goals, completed work, blockers, decisions, sentiment arc, recurring patterns, and recommendations |
93
- | **Smart auto-trigger** | Reports generate automatically when sessions hit ≥3 tool errors, ≥2 blockers, &gt;60min, abandoned plan steps, or sustained frustration |
94
- | **Pattern memory loop** | Recurring patterns from post-mortems are stored as `pattern` memories — your AI learns from its own session history |
95
- | **`/observe` dashboard** | Live session stats: tool calls, errors, files changed, blockers, milestones, topic shifts |
96
- | **`/postmortem` commands** | Generate now, view last, list all, or `--since 7d` for cross-session trend analysis |
97
- | **Eval reliability fix** | `/eval report` now uses `fs.existsSync` instead of an ANSI-sensitive prefix check |
98
- | **CI dependency stability** | Switched `npm ci` → `npm install` to handle platform-specific native deps |
91
+ | **Skill crystallization** | Post-mortems identify reusable procedures → opt-in prompt → saved as auto-triggering skills |
92
+ | **Runtime trigger matching** | Crystallized skills merge with hardcoded skill triggers — keyword match fires both |
93
+ | **`/skills crystallize`** | Manually crystallize skills from the most recent post-mortem JSON sidecar |
94
+ | **`/skills list --auto`** | List all crystallized skills with dates, confidence, and trigger keywords |
95
+ | **Post-mortem JSON sidecar** | Every post-mortem now writes a `.json` alongside the `.md` for lossless re-parsing |
96
+ | **Extended LLM schema** | Post-mortem prompt includes crystallization rules — narrow, reusable, confidence-gated |
99
97
 
100
98
  <details>
101
99
  <summary><strong>Highlights from earlier releases</strong></summary>
102
100
 
101
+ **v0.24 — Observation & post-mortem**
102
+ - Passive session telemetry (tool calls, errors, file changes, sentiment, blockers)
103
+ - LLM-generated post-mortem reports with smart auto-trigger
104
+ - Pattern memory loop — the agent learns from its own session history
105
+ - `/observe` dashboard, `/postmortem` commands
106
+
103
107
  **v0.18 — User onboarding**
104
108
  - Interactive first-run setup capturing name, role, expertise, communication style
105
109
  - 13 showcase templates (fitness, freelancer, Muslim, finance, etc.)
@@ -1044,7 +1048,7 @@ On session end, if any smart trigger fires (≥3 tool errors, ≥2 blockers, &gt
1044
1048
  | `/rules` | View guardrails `[add\|remove\|toggle ...]` |
1045
1049
  | `/workflows` | View workflows `[add\|remove ...]` |
1046
1050
  | `/tools` | View tools `[add\|remove ...]` |
1047
- | `/skills` | View skills `[install\|uninstall ...]` |
1051
+ | `/skills` | View skills `[install\|uninstall\|crystallize\|list --auto]` |
1048
1052
  | `/eval` | View evaluation `[milestone ...]` |
1049
1053
  | `/memory` | View memories `[search\|clear\|timeline]` |
1050
1054
  | `/observe` | Live session telemetry dashboard `[pause\|resume]` |