feat(term): full-screen terminal console for agents, sessions and cost

README.md

@@ -88,6 +88,7 @@ For the Skill, curl, and git-clone alternatives, plus PEP 668 systems and secret
 - ⚖️ [Layered Policy & Exemptions](docs/policy-layers-and-exemptions.md) covers per-rule observe/enforce, the non-overridable floor, and admin-granted, time-boxed exemptions across org / project / repo layers
 - 📡 [Live Telemetry](docs/live-telemetry.md) covers the optional enterprise control-plane link — device enrollment, signed remote policy, and redacted telemetry streamed to a self-hosted org dashboard
 - 📊 [Dashboard](docs/dashboard.md) covers the terminal and local web dashboards plus session forensics
+- 🖥️ [Terminal Console](docs/terminal-console.md) is `prismor term` — a full-screen agent → session tree with a live event tail, policy precedence, and per-session token cost, without leaving the terminal
 - 🧾 [Signed Audit Trail](docs/audit-trail.md) hash-chains and Ed25519-signs every agent action locally, so `prismor trail verify` proves the history hasn't been edited, deleted, or rewritten
 - 📑 [Attestation Bundle](docs/attestation-bundle.md) packages posture, agent inventory, host discovery, framework-control coverage (OWASP LLM/Agentic, NIST AI RMF, EU AI Act), and the trail anchor into one Ed25519-signed file an auditor re-verifies with `prismor attest verify`
 - 🔦 [Host Discovery](docs/attestation-bundle.md#host-discovery) sweeps the machine with `prismor discover` and flags any AI agent running without Prismor hooks (shadow AI)

docs/cli-reference.md

@@ -58,6 +58,7 @@ prismor
 │   ├─ deps                   Check project deps vs. threat feed
 │   ├─ analyze / ingest       Run the engine over a JSONL session
 │   ├─ sessions / session     List / show stored sessions
+│   ├─ term                   Full-screen console (agents, sessions, live tail)
 │   ├─ trail <action>         verify · show · checkpoint — signed audit trail
 │   ├─ attest [verify|coverage]  Signed evidence bundle + framework coverage
 │   ├─ discover               Sweep host for ungoverned AI agents (shadow AI)
@@ -160,6 +161,7 @@ Full policy model, rule schema, and the default rule list: [Prismor](prismor-run
 | `prismor attest coverage` | `--json`, `--workspace` | Show which compliance-framework controls the active policy covers (OWASP LLM/Agentic, NIST AI RMF, EU AI Act). |
 | `prismor discover` | `--json`, `--workspace` | Sweep this host for AI agents and flag any running without Prismor hooks (shadow AI). Host-local, read-only. See [Host discovery](attestation-bundle.md#host-discovery). |
 | `prismor status --all` | `--days N` | Terminal overview of every registered workspace. See [Dashboard](dashboard.md). |
+| `prismor term` | — | Full-screen terminal console: agent → session tree, live event tail, policy precedence, per-session token cost. Falls back to a plain table when stdout isn't a tty. See [Terminal Console](terminal-console.md). |
 | `prismor dashboard` | `--port`, `--host`, `--no-open` | Local web dashboard at `http://127.0.0.1:7070` (opens a browser tab). See [Dashboard](dashboard.md). |
 | `prismor serve` | `--port`, `--host`, `--no-open` | _Deprecated_ alias of `dashboard --no-open` (headless server only). |
 
@@ -268,5 +270,6 @@ Scoring table, IOC feed, ecosystem support: [Supply Chain](supply-chain.md).
 - [Skill Scanner](skill-scanner.md) — MCP + skill risk scanning
 - [Sweep & Cloak](sweep-and-cloak.md) — secret prevention
 - [Semantic Guard](semantic-guard.md) — LLM-assisted injection defense
+- [Terminal Console](terminal-console.md) — `prismor term`, the full-screen agent/session console
 - [Canary](canary.md) · [IAM](iam.md) · [Scoped Agent](scoped-agent.md) · [Learning](learning.md) · [Dashboard](dashboard.md)
 - [Docker & Containers](docker.md) · [Architecture](architecture.md)

docs/dashboard.md

@@ -159,6 +159,7 @@ viewer, with full rule metadata.
 
 ## See also
 
+- [Terminal Console](terminal-console.md) — `prismor term`, a full-screen console over this same store: agent → session tree, live event tail, per-session cost
 - [Prismor](prismor-runtime.md) — session-log schema and the audit command
 - [Learning](learning.md) — mines this same history for new rules
 - [CLI Reference](cli-reference.md) — all commands at a glance

docs/terminal-console.md

@@ -0,0 +1,181 @@
+# Terminal Console (`prismor term`)
+
+`prismor term` is a full-screen console for everything Prismor has recorded: a
+tree of your agents and their sessions, a live event tail, the policy layer in
+effect, and what each session cost in tokens.
+
+It is a **renderer only**. Every number comes from
+[`prismor/runtime/store.py`](../prismor/runtime/store.py) — the same data layer
+behind [`prismor dashboard`](dashboard.md) — which aggregates across every
+registered workspace and re-cloaks secrets on read, so evidence panes show
+`@@SECRET:NAME@@` placeholders and never real values.
+
+Implementation: [`prismor/runtime/term.py`](../prismor/runtime/term.py), cost in
+[`prismor/runtime/cost.py`](../prismor/runtime/cost.py).
+
+```bash
+prismor term
+```
+
+Needs a real terminal. Piped or redirected (CI, `| head`, no tty), it prints a
+plain agent + session table instead, so it stays scriptable.
+
+---
+
+## The layout
+
+```
+ >_ PRISMOR  TERM  │ Org: acme  │ Policy: default  │ Agents: 35  │ prices live  │ FOLLOW
+ Agents ▸ sessions  ↓last run    │  Detail
+────────────────────────────────│──────────────────────────────────────────────
+   All agents                   │  Session    d981011d-6f54-4516-a698-2afe8db…
+ ▾ ○ claude            512/229  │  Agent      claude  (claude)
+     6a5d59a6-e  r90 2f   — 1d  │  Risk       70/100 · 1 findings
+     d981011d-6  r70 1f $426 2d │  Immunity   active
+     19d34f10-6  r0 0f $0.05 1d │  Cost       $426.00 est · 1629 turns · claude-opus-5
+ ▸ ○ codex              64/32   │  Tokens     in 3.1k · out 1.4M · cache r687.6M/w7.4M
+ ▸ ○ langchain          11/11   │
+                                │  Events  [page 1/3 of 60 · session · filter: all]
+                                │───────────────────────────────────────────────
+                                │  23:08:18 claude  BLOCK shell: rm -rf /
+                                │  23:08:01 claude  allow prompt
+ [j/k] Move  [→/←] Expand  [Tab] Pane  [[/]] Page  [Enter] Detail  [s] Sort  …
+```
+
+**Header** — org (or `local` when unenrolled), the winning policy layer, agent
+count, token-price freshness, and whether the tail is following.
+
+**Left pane** — a two-level tree. Agents show `total/flagged` call counts and a
+status dot. Expand one and its sessions appear beneath, each with risk,
+findings, cost, and age: `r90 2f $426 2d`.
+
+**Right pane** — detail for whatever is selected, above an event tail scoped to
+the same thing: all agents, one agent, or one session.
+
+---
+
+## Keys
+
+| Key | Does |
+|---|---|
+| `j` / `k`, `↑` / `↓` | Move in the focused pane |
+| `→` / `l`, `←` / `h` | Expand / collapse an agent's sessions |
+| `Tab` | Switch focus between the tree and the event tail |
+| `Enter` | On an agent: expand. On an event: full detail (rule, category, evidence) |
+| `[` / `]`, `PgUp` / `PgDn` | Page the event tail |
+| `g` / `G` | Jump to top / bottom |
+| `s` | Cycle session sort: last run → risk → findings → cost |
+| `f` | Toggle follow (live tail, refreshes every 2s) |
+| `v` | Cycle verdict filter: all → blocked → allowed |
+| `p` | Policy precedence overlay |
+| `P` | Pause / resume immunity for the selected session (asks first) |
+| `R` | Resume the selected session in Claude Code (asks first) |
+| `r` | Refresh everything, including token prices |
+| `q` | Quit |
+
+---
+
+## Sessions
+
+Expanding an agent loads its sessions newest-first. Two limits are worth knowing
+because the UI states them rather than hiding them:
+
+- At most **50 sessions per agent** are loaded (scanning up to 5 pages). When
+  that bites, the tree shows `… showing first 50`. An agent with 512 sessions
+  gives you the 50 most recent, not a random 50.
+- `s` re-sorts **what is loaded**, not the full history. Sorting by cost ranks
+  within those 50 most-recent sessions.
+
+Selecting a session scopes the event tail to it and shows its risk, workspace,
+scoped rules, and immunity state.
+
+### Pausing a session
+
+`P` toggles immunity for one session via the same control the web dashboard
+uses. It confirms first, and it writes: paused sessions stop being screened
+until you resume them. It affects that session only — not the agent, not the
+workspace.
+
+### Resuming a session in Claude Code
+
+`R` hands the terminal to `claude --resume <session-id>`, run in that session's
+original working directory. `prismor term` restores itself when you exit claude.
+
+This works because Claude Code's conversation ids *are* the session ids Prismor
+records. It refuses, with the reason, when the session ran under a different
+framework, when the `claude` CLI isn't on `PATH`, or when the recorded workspace
+no longer exists (common for sessions that ran in temp directories).
+
+---
+
+## Cost
+
+Prismor's event store records *what* an agent did, never how many tokens it
+took — there is no usage column and no usage payload in any event. So cost is
+joined from two outside sources:
+
+| | Source |
+|---|---|
+| **Prices** | The published feed at `https://www.aipricing.guru/api/pricing.json`, cached to `~/.prismor/pricing-cache.json` for 12h |
+| **Usage** | The agent's own transcript — Claude Code writes `~/.claude/projects/<slug>/<session-id>.jsonl` with a `usage` block per turn |
+
+The header reports which prices are in play: `prices live` (just fetched),
+`prices cached` (within TTL), `prices STALE` (fetch failed, serving the cache),
+or `prices n/a` (no network and no cache — costs render unpriced). Nothing
+blocks on the network; the fetch has a 6s timeout and falls back.
+
+**Read the numbers with these caveats**, all of which the UI marks:
+
+- Cost is shown only for agents that keep transcripts — Claude Code today.
+  Other frameworks show `—` (unknown), never `$0.00`. A `·` means *not priced
+  yet*; it fills in within a second.
+- Agent-level totals say `across N of M loaded sessions`. They are not a
+  lifetime total for that agent.
+- The price feed publishes no **cache-write** rate, so that component uses
+  Anthropic's standard 1.25x input multiplier. That is the 5-minute-TTL figure;
+  1-hour-TTL caching bills 2x, so long-cache sessions are underestimated.
+  Everything is labelled `est` for this reason.
+- These are **list-price API costs**. On a subscription plan your real marginal
+  cost was the flat fee — the figure answers "what would this have cost on the
+  API", which is useful for attribution, not a bill.
+
+---
+
+## Event paging
+
+The tail fetches exactly one screenful at a time, so paging stays fast on a
+large store.
+
+Counts are reported honestly, and they differ by scope:
+
+- **Session-scoped** events are read in full (the store returns up to 60) and
+  sliced locally, so the count is exact: `page 1/3 of 60`.
+- **Agent- and all-scoped** events come from a store query whose reported total
+  grows as you page deeper — it is a floor, not a count. The header says
+  `page 3 · 23 of 207+` and `]` simply stops when a page comes back short.
+
+---
+
+## Performance notes
+
+The console is built to never block on a keypress:
+
+- Navigation redraws immediately from cached rows and re-queries only once the
+  selection has been still for ~120ms, so holding `j` never queues queries.
+- The 24-hour aggregate (the "All agents" KPI panel) is the one expensive query
+  in the app. It is never run on startup or during a redraw — the panel paints
+  as `computing 24h totals…` and fills in when idle.
+- Session pricing runs in small batches while idle.
+- The screen repaints only when something changed.
+
+If the terminal is smaller than 50x10 the console says so rather than drawing a
+broken frame.
+
+---
+
+## See also
+
+- [Dashboard & Sessions](dashboard.md) — the web dashboard and session forensics over the same store
+- [Policy Layers & Exemptions](policy-layers-and-exemptions.md) — what the `p` overlay is showing
+- [Scoped Agent](scoped-agent.md) — the per-session controls `P` toggles
+- [CLI Reference](cli-reference.md) — all commands at a glance

prismor/runtime/cli.py

@@ -955,6 +955,18 @@ def main(argv: Optional[List[str]] = None) -> None:
             _print_status_overview(workspace)
         return
 
+    # ── term: full-screen terminal console (curses) ─────────────────────
+    if args.command == "term":
+        from prismor.runtime.term import run_term
+        registered = list_registered_workspaces()
+        if not registered:
+            sys.stderr.write(
+                "[prismor] Warning: no registered workspaces found.\n"
+                "         Run 'prismor install-hooks' in a project first to collect data.\n"
+            )
+        run_term()
+        return
+
     # ── analyze ────────────────────────────────────────────────────────
     if args.command == "analyze":
         # Accept `analyze <file>` as shorthand for `analyze --input <file>`.
@@ -2532,6 +2544,12 @@ def build_parser() -> argparse.ArgumentParser:
         help="With --all: show activity for the last N days (default: 7)",
     )
 
+    # ── term ───────────────────────────────────────────────────────────
+    subparsers.add_parser(
+        "term",
+        help="Full-screen terminal console: agents, sessions, live event tail",
+    )
+
     # ── analyze ────────────────────────────────────────────────────────
     analyze = subparsers.add_parser("analyze", help="Analyze a session (or current session if no --input)")
     analyze.add_argument("file", nargs="?", help="Path to JSONL session file (same as --input). If omitted, analyzes most recent session")