📜 omparc (omptale)

Status: planned. Pick: omparc (plugin id) · display omptale · sole human command /arc.
Hub: 🔌 OMP Extensions · Contract: 🧭 Extension UX Contract · Hermes ref: ~/.hermes/hermes-agent/agent/learning_graph.py

1. 💎 Why this name

Candidate	Verdict	Why
omparc ✅	Chosen	Learning arc — timeline + graph; short; /arc; no clash with ompfuse traces
omptale	Tagline	Narrative hook — “your agent’s learning tale”
ompthrive	Alt	Growth vibe; less visual/timeline-specific
ompersist	❌	Sounds like persistence layer only (ompkeep already owns that)
ompmneme	Alt	Nerdy tie to mnemopi; harder to say
omp_gnosis	❌	Too mystical; overlaps skill/guru tone

Not ompjourney — Hermes owns the word; omparc is OMP-native with compatible UX.

2. 🎯 What Hermes /journey does (enticing baseline)

Three slash aliases → one overlay:

/journey  |  /learning  |  /memory-graph

Data model (build_learning_graph()):

Node kind	Source	Filter
skill ●	SKILL.md under profile + base	Non-base AND (agent-created OR use_count > 0)
memory ◆	MEMORY.md / USER.md chunks (§ separators)	All chunks surfaced

Edges: related_skills (declared) + lexical memory↔skill overlap (top 4 per card).

UI: Desktop starmap constellation · TUI timeline bars (oldest→newest) · age gradient ink · CRUD (learning.edit / delete / detail).

One-liner: Learning made visible — not just storage, but how the agent grew over time.

3. 🧠 What OMP already has (build on, don’t duplicate)

Your ~/.omp/agent/config.yml today:

autolearn:
  enabled: true
  autoContinue: true
memory:
  backend: mnemopi
advisor:
  enabled: true
  syncBacklog: "1"

Layer	System	Role for omparc
Retain/recall	mnemopi SQLite banks	Memory ◆ nodes + timestamps
Governed memory	ompkeep L1/L2 JSONL + inbox	High-trust ◆ nodes + patch history
Skills	~/.omp/agent/managed-skills/ + project skills	Skill ● nodes + usage
Autolearn	autolearn.* config	Future: auto-skill proposals → arc events
Advisor	advisor.syncBacklog	Backlog items as arc milestones

Gap: OMP has memory + skills + autolearn but no unified visual journey — that’s omparc.

4. 💎 Blueprint

id: omparc
display: omptale
path: ~/.omp/agent/extensions/omparc.ts
status: planned
human_command: /arc
command_owner: omparc
human_ui: "overview → lane → drill-down; no registered aliases"
widget: pi.setWidget("omparc:timeline") | TUI overlay
data_sources:
  skills: managed-skills + project .omp/skills + usage from sessions
  memory_mnemopi: mnemopi bank stats + /memory view export
  memory_ompkeep: L2.playbooks + reinforcement timestamps
  sessions: stats.db message counts (optional density)
graph:
  nodes: [skill, memory, milestone]
  edges: [related_skills, lexical_overlap, retain_link, skill_created]
render:
  terminal: timeline frames (port learning_graph_render.py ideas)
  widget: compact sparkline + node list (phase 1)
mutations: []
runtime: omp only
deps: "@oh-my-pi/pi-tui", optional ompkeep read-only API
scope: "read-only MVP; drill-down delegates to the owning system"

5. 🏗️ Architecture

flowchart TB
  subgraph sources["Data sources"]
    SK[Skills + usage]
    MN[mnemopi banks]
    OK[ompkeep L2/L1]
    ST[stats.db optional]
  end
  subgraph omparc["omparc extension"]
    G[build_learning_graph]
    R[render timeline / widget]
    C[/arc command]
  end
  subgraph ui["OMP TUI"]
    W[Widget overlay]
    T[Terminal frames]
  end
  SK --> G
  MN --> G
  OK --> G
  ST --> G
  G --> R
  C --> R
  R --> W
  R --> T

6. 📋 Implementation phases

Phase	Deliverable	Effort
P0	buildOmpLearningGraph() — skills + mnemopi recall metadata	Read-only graph JSON
P1	/arc TUI — ASCII timeline (Hermes learning_graph_render port)	Slash command
P2	Widget mode above editor (like ompbar)	setWidget
P3	Node drill-down → /memory view, skill path open	Mutations delegate
P4	Autolearn events on arc (new skill retained → ◆ on timeline)	Hook agent_end

OmpKeep lane

OmpKeep is the governed-memory lane, not a second memory store. Render only safe aggregate/change metadata by default:

lane: governed_memory
source: "public persistent-intelligence read adapter"
nodes: [candidate, patch, approved_l2, reinforcement, deprecation, tombstone]
default_detail: "counts, timestamps, scope, and status"
private_content: "explicit drill-down only; obey record scope/status"
mutations: "none — /arc delegates to /ok"

The MVP must not import private components from ompkeep.ts, edit its JSONL, or duplicate OmpKeep’s graph/timeline commands.

7. 🔌 OMP extension sketch

// ~/.omp/agent/extensions/omparc.ts — skeleton
export default function omparc(pi: ExtensionAPI) {
  pi.setLabel("omptale");
 
  pi.registerCommand("arc", { /* status-first menu + focused lanes */ });
  pi.on("session_start", async (_, ctx) => {
    // optional: arm widget like ompbar
  });
  // buildGraph(): merge skill scan + mnemopi stats + ompkeep L2 timestamps
}

Reuse from Hermes (Python → TS port, same contracts):

• build_skill_nodes + filter learned signal
• _memory_cards chunking pattern (OMP: mnemopi export or ompkeep statements)
• _memory_skill_edges lexical scoring
• render_frames timeline layout

8. 🆚 Hermes vs omparc

	Hermes /journey	omparc /arc
Memory	MEMORY.md prose chunks	mnemopi + ompkeep + optional markdown
Skills	~/.hermes/skills	managed-skills + project skills
Graph UI	Desktop starmap + TUI timeline	TUI timeline → widget (phased)
CRUD	edit/delete nodes	Read-only MVP; delegate to owning system
Telemetry	separate	Langfuse via ompfuse

9. ✅ Done when (MVP)

• /arc opens timeline in omp TUI
• Shows ≥1 skill ● and ≥1 memory ◆ from real data
• Edges between related skills visible
• OmpKeep lane shows governed-memory lifecycle summaries without text leakage
• No write path yet — read-only is fine for v0

10. 🧪 Smoke (post-build)

/arc
/memory stats
# create/use a skill → re-run /arc → new ● appears

11. 🔗 Links

Hermes learning graph	~/.hermes/hermes-agent/agent/learning_graph.py
OMP ↔ Hermes memory bridge	omp.0rk.de
ompbar	Shared widget patterns
ompfuse	Trace retention alongside learning arc