-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathcli-reference.html
More file actions
81 lines (81 loc) · 7.78 KB
/
Copy pathcli-reference.html
File metadata and controls
81 lines (81 loc) · 7.78 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
<!doctype html>
<html lang="en">
<head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>Invart CLI Reference</title><link rel="stylesheet" href="style.css"></head>
<body>
<header class="hero"><span class="badge">CLI</span><h1>The command groups users need first.</h1><p>Run <code>invart --help</code> for the complete parser. These are the entry points most people should start with.</p><nav class="nav"><a href="index.html">Docs Home</a><a href="five-layer-operator-guide.html">Operate L1-L5</a><a href="api-sdk.html">API & SDK</a><a href="evaluation.html">Evaluation</a></nav></header>
<main class="wrap">
<section>
<h2>User intent to command</h2>
<table>
<tr><th>I want to...</th><th>Start with</th></tr>
<tr><td>Scan before an agent runs</td><td><code>invart pre-runtime --target . --save</code></td></tr>
<tr><td>Run one managed session</td><td><code>invart run --target . --agent codex --goal "..." -- <command></code></td></tr>
<tr><td>Inspect L1-L5 for a ledger</td><td><code>invart runtime layers --ledger ledger.jsonl --out-dir .invart/layers</code></td></tr>
<tr><td>Explain a risky path</td><td><code>invart policy check-path --ledger ledger.jsonl --out path-policy.json</code></td></tr>
<tr><td>Inspect mediation state</td><td><code>invart mediation inspect --ledger ledger.jsonl</code></td></tr>
<tr><td>Export reviewable evidence</td><td><code>invart evidence export --ledger ledger.jsonl --out-dir .invart/evidence</code></td></tr>
<tr><td>Validate real-agent integration</td><td><code>invart real-agent check --agent claude-code --out-dir .invart/real-agent</code></td></tr>
<tr><td>Run task-shaped installed-agent slice</td><td><code>invart experiment task-agent --agent claude-code --agent codex --require-installed --out-dir .invart/task-agent</code></td></tr>
<tr><td>Run product benchmarks</td><td><code>invart eval benchmark --suite full-product-readiness</code></td></tr>
</table>
</section>
<section class="grid">
<div class="card"><h3>Pre-runtime</h3><pre>invart pre-runtime --target . --save</pre></div>
<div class="card"><h3>Managed session</h3><pre>invart session start --target . --agent codex --goal "..."
invart run --target . --agent codex --goal "..." -- <command></pre></div>
<div class="card"><h3>Runtime event</h3><pre>invart runtime analyze-event --event '{"type":"shell","command":"rm -rf ."}'
invart runtime shell --session demo --ledger .invart/demo.jsonl -- <command>
invart runtime layers --ledger .invart/demo.jsonl --out-dir .invart/layers</pre><p><code>runtime layers</code> exports a L1-L5 operation workflow with proof, replay, path graph, coverage, audit, and evidence manifest links.</p></div>
<div class="card"><h3>Proof and gate</h3><pre>invart proof export --ledger ledger.jsonl --out proof.json
invart proof verify --proof proof.json --ledger ledger.jsonl
invart gate verify --proof proof.json --ledger ledger.jsonl --mode ci</pre></div>
<div class="card"><h3>Replay and audit</h3><pre>invart replay export --ledger ledger.jsonl --out replay.html
invart audit report --ledger ledger.jsonl --out-dir .invart/audit</pre></div>
<div class="card"><h3>Evidence workspace</h3><pre>invart evidence export --ledger ledger.jsonl --out-dir .invart/evidence
invart evidence verify --bundle .invart/evidence/manifest.json
invart evidence inspect \
--manifest .invart/evidence/manifest.json \
--out-dir .invart/evidence-workspace \
--require-layer-workflow</pre><p><code>evidence inspect</code> verifies bundle hashes and required artifacts, then reports whether the L5 review workspace can answer who, what, why, policy, approval, outcome, and coverage. Optional layer workflow and adapter package requirements turn missing links into gate failures.</p></div>
<div class="card"><h3>Real agent conformance</h3><pre>invart adapter profile --kind claude-code
invart adapter profiles
invart adapter profiles --track managed_wrapper
invart real-agent check --agent claude-code --out-dir .invart/real-agent
invart real-agent run --agent claude-code --require-live --out-dir .invart/live-claude -- <claude-or-fixture-command>
invart real-agent run --agent opencode --require-live --out-dir .invart/live-opencode -- <opencode-or-fixture-command>
invart real-agent run --agent gemini-cli --require-live --out-dir .invart/live-gemini -- <gemini-or-fixture-command>
invart real-agent run --agent aider --require-live --out-dir .invart/live-aider -- <aider-or-fixture-command>
invart real-agent run --agent codex --require-live --out-dir .invart/live-codex -- <codex-or-fixture-command>
invart real-agent report --run-dir .invart/real-agent --out .invart/real-agent/report.html</pre><p>Use <code>--require-live</code> when missing local binaries should fail instead of being recorded as blocked evidence. Fixture-backed runs validate the Invart adapter contract; path-resolved probes validate that an installed local binary can enter the wrapper, but they are not provider-task or benchmark validation. <code>adapter profiles</code> lists priority tracks and keeps vendor/cloud import separate from Invart-mediated control. <code>real-agent check</code> emits a conformance contract row so imported, discovered, fixture-backed, path-resolved, vendor-owned, and missing-binary evidence cannot be mixed into stronger claims.</p></div>
<div class="card"><h3>Task-shaped installed-agent experiment</h3><pre>invart experiment task-agent \
--agent claude-code \
--agent codex \
--require-installed \
--out-dir .invart/task-agent</pre><p><code>experiment task-agent</code> runs the v0.53 deterministic local task-shaped slice for installed agent products. It resolves requested binaries, runs benign, destructive-token, credential-exposure-token, and critical-token commands through Invart-managed wrappers, and writes JSON/HTML reports with direct-baseline and managed-wrapper outcomes. This is local managed-wrapper path-governance evidence, not proof that the provider model solved the task or an upstream benchmark score.</p></div>
<div class="card"><h3>Claude Code reference adapter</h3><pre>invart adapter claude-code \
--target . \
--out-dir .invart/claude-reference \
--binary claude \
--require-live \
--hook-events .invart/claude-hooks.jsonl \
--policy-mode managed \
-- <claude-or-harness-command></pre><p>Records Claude-style hook events, mediates the child command, and exports ledger, proof, replay, path graph, coverage, audit, evidence manifest, layer-runtime workflow, and evidence workspace. <code>--require-live</code> fails when the Claude binary is missing. Binary-shaped fixtures verify the adapter path but are not reported as vendor-installed live runs. Portable subprocess supervision is marked as degraded process-tree coverage unless native supervision is enabled.</p></div>
<div class="card"><h3>Evaluation</h3><pre>invart eval list
invart eval benchmark --suite full-product-readiness
invart eval benchmark --suite v0.9.3-agent-adapter-contract
invart eval benchmark --suite v0.9.4-claude-reference-adapter
invart eval benchmark --suite v0.9.5-priority-agent-tracks
invart eval benchmark --suite v0.9.6-layer-runtime-workflow
invart eval benchmark --suite v0.9.7-evidence-workspace-gate
invart eval benchmark --suite v0.9.8-claude-full-live-adapter
invart eval benchmark --suite v0.9.9-conformance-contract-v2
invart eval benchmark --suite v0.9.10-opencode-real-adapter
invart eval benchmark --suite v0.9.11-terminal-agent-managed-wrappers
invart eval benchmark --suite v0.9.12-codex-boundary
invart eval benchmark --suite v0.9.13-ide-bridge-inventory
invart eval benchmark --suite v0.53-task-agent-installed-slice
invart roadmap status --require-full</pre></div>
</section>
</main>
</body>
</html>