You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
(Most other flags — --agent-image, --agent-timeout, --audit-dir, --config, --copilot-api-target, --diagnostic-logs, --difc-proxy-*, --docker-host, --enable-dind, --enable-dlp, --enable-host-access, --env-*, --exclude-env, --gemini-api-*, --image-registry, --image-tag, --log-level, --memory-limit, --no-rate-limit, --openai-api-*, --proxy-logs-dir, --rate-limit-*, --ruleset-file, --session-state-dir, --skip-pull, --ssl-bump, --tty, --upstream-proxy, --work-dir, -b/--build-local, -e/--env, -k/--keep-containers, -v/--mount — are consistently documented in usage.md and cli-reference.md, which is expected as the two detailed reference sources; README.md/CLAUDE.md intentionally stay high-level.)
⚠️ Issues Found
1. 🔴 CRITICAL: --enable-api-proxy deprecation is not reflected in any documentation
Location: src/commands/validators/security-mode.ts (lines ~19, 128–148) is the source of truth.
Issue: The API proxy sidecar is now always enabled for credential isolation. Passing --enable-api-proxy triggers a deprecation warning (⚠️ --enable-api-proxy is deprecated and no longer needed. The API proxy is always enabled.). Passing --no-enable-api-proxy is now a hard error (process.exit(1), message: ❌ --no-enable-api-proxy is not allowed. The API proxy is always enabled for credential isolation.).
Expected: Docs should state the API proxy is on by default, mark --enable-api-proxy as a deprecated no-op flag, and document that --no-enable-api-proxy will abort execution.
Current:
docs/usage.md (~lines 96–120): describes --enable-api-proxy as opt-in (default: false) and marks ~6 dependent flags as "(requires --enable-api-proxy)".
docs-site/src/content/docs/reference/cli-reference.md (line ~59, and ~15 example commands throughout, e.g. line 726): lists it in the flag table as false default and shows sudo -E awf --enable-api-proxy ... in numerous examples as if required.
CLAUDE.md/AGENTS.md (architecture section, "API Proxy Sidecar (optional)"): states "Enabled via --enable-api-proxy; not started otherwise" — directly contradicts current always-on behavior.
README.md: silent on the flag, doesn't clarify always-on status either.
Suggestion: Update usage.md, cli-reference.md, and CLAUDE.md/AGENTS.md to (a) state the API proxy is always enabled, (b) mark --enable-api-proxy as deprecated/no-op with a warning, (c) document --no-enable-api-proxy as removed/hard error, (d) remove "(requires --enable-api-proxy)" qualifiers from dependent flag descriptions, and (e) drop --enable-api-proxy from example commands in cli-reference.md.
2. Entire Cloud Hypervisor flag family undocumented
Issue: None of these appear in docs/usage.md or cli-reference.md. Only a high-level architectural mention exists in CLAUDE.md (linking to docs/cloud-hypervisor-foundation.md).
Suggestion: Add a dedicated "Cloud Hypervisor (preview)" section to usage.md and cli-reference.md, or explicitly cross-reference docs/cloud-hypervisor-foundation.md from both.
3. Token-steering/budget flags undocumented or partially documented
Location: src/cli-options.ts lines ~415–447.
Issue: --enable-token-steering is in cli-reference.md but missing from usage.md. Its companion flags --max-cache-misses, --max-model-multiplier, --max-model-multiplier-cap, and --max-permission-denied are missing from all docs.
Suggestion: Add these four flags to both usage.md and cli-reference.md alongside --enable-token-steering, and add --enable-token-steering itself to usage.md.
Issue: Gemini/OpenAI/Anthropic equivalents (--gemini-api-target, --openai-api-target, --anthropic-api-target, etc.) are documented in both usage.md and cli-reference.md, but the Vertex AI variants are absent from all docs.
Suggestion: Add Vertex AI flags alongside the other provider-target flags for consistency.
5. Cross-doc drift between usage.md and cli-reference.md
Issue: Several flags appear in one detailed doc but not the other: --anthropic-auto-cache, --anthropic-cache-tail-ttl, --docker-host-path-prefix, --network-isolation, --pids-limit, --topology-attach (all in usage.md, missing from cli-reference.md); --enable-token-steering (in cli-reference.md, missing from usage.md).
Suggestion: Reconcile the two reference docs so each carries the complete flag set; consider generating one canonical flag table to avoid future drift.
6. --reflect documented only in README.md
Issue: --reflect appears in README.md but not in the two detailed reference docs (usage.md, cli-reference.md), the reverse of the usual gap pattern.
Suggestion: Add --reflect to usage.md and cli-reference.md with a description matching README's usage.
7. Minor undocumented flags
--container-runtime, --no-network-isolation, --anthropic-api-auth-header, --openai-api-auth-header are absent from all docs.
Suggestion: Low-priority addition to usage.md/cli-reference.md flag tables.
📋 Recommendations
High Priority: Fix the --enable-api-proxy documentation across docs/usage.md, cli-reference.md, and CLAUDE.md/AGENTS.md — this is actively misleading since it describes opt-in behavior that no longer exists and doesn't warn about the --no-enable-api-proxy hard error.
Medium Priority: Document the Cloud Hypervisor flag family (11 flags) and the token-steering budget flags (--max-cache-misses, --max-model-multiplier, --max-model-multiplier-cap, --max-permission-denied), and the Vertex AI target/base-path flags.
Medium Priority: Reconcile drift between usage.md and cli-reference.md so both carry the same flag set (6+ flags differ between them).
Low Priority: Add --reflect to the detailed reference docs; document --container-runtime, --no-network-isolation, and the *-api-auth-header flags.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
📊 Summary
src/cli-options.tsimplementation: ~82 (including short-form aliases and the 5logs/predownloadsubcommands)--enable-api-proxydeprecation is not reflected in any documentation file — this is the most impactful finding.✅ Flags Status
-d, --allow-domains--allow-domains-file--allow-urls--allow-host-ports/--allow-host-service-ports--block-domains/--block-domains-file--dns-servers--dns-over-https--enable-api-proxy--no-enable-api-proxy--enable-token-steering--max-cache-misses--max-model-multiplier--max-model-multiplier-cap--max-permission-denied--anthropic-auto-cache--anthropic-cache-tail-ttl--anthropic-api-auth-header/--openai-api-auth-header--vertex-api-target/--vertex-api-base-path--container-runtime--no-network-isolation--network-isolationonly in usage.md)--cloud-hypervisor-*(11 flags: binary/kernel/rootfs/supervisor/vcpus/memory-mib/api-timeout-ms/preview + 5-sha256variants)--docker-host-path-prefix--network-isolation--pids-limit--topology-attach--reflectlogs,logs stats,logs summary,logs audit,predownload(Most other flags —
--agent-image,--agent-timeout,--audit-dir,--config,--copilot-api-target,--diagnostic-logs,--difc-proxy-*,--docker-host,--enable-dind,--enable-dlp,--enable-host-access,--env-*,--exclude-env,--gemini-api-*,--image-registry,--image-tag,--log-level,--memory-limit,--no-rate-limit,--openai-api-*,--proxy-logs-dir,--rate-limit-*,--ruleset-file,--session-state-dir,--skip-pull,--ssl-bump,--tty,--upstream-proxy,--work-dir,-b/--build-local,-e/--env,-k/--keep-containers,-v/--mount— are consistently documented in usage.md and cli-reference.md, which is expected as the two detailed reference sources; README.md/CLAUDE.md intentionally stay high-level.)1. 🔴 CRITICAL:
--enable-api-proxydeprecation is not reflected in any documentationsrc/commands/validators/security-mode.ts(lines ~19, 128–148) is the source of truth.--enable-api-proxytriggers a deprecation warning (⚠️ --enable-api-proxy is deprecated and no longer needed. The API proxy is always enabled.). Passing--no-enable-api-proxyis now a hard error (process.exit(1), message:❌ --no-enable-api-proxy is not allowed. The API proxy is always enabled for credential isolation.).--enable-api-proxyas a deprecated no-op flag, and document that--no-enable-api-proxywill abort execution.docs/usage.md(~lines 96–120): describes--enable-api-proxyas opt-in (default: false) and marks ~6 dependent flags as "(requires --enable-api-proxy)".docs-site/src/content/docs/reference/cli-reference.md(line ~59, and ~15 example commands throughout, e.g. line 726): lists it in the flag table asfalsedefault and showssudo -E awf --enable-api-proxy ...in numerous examples as if required.CLAUDE.md/AGENTS.md(architecture section, "API Proxy Sidecar (optional)"): states "Enabled via--enable-api-proxy; not started otherwise" — directly contradicts current always-on behavior.README.md: silent on the flag, doesn't clarify always-on status either.--enable-api-proxyas deprecated/no-op with a warning, (c) document--no-enable-api-proxyas removed/hard error, (d) remove "(requires --enable-api-proxy)" qualifiers from dependent flag descriptions, and (e) drop--enable-api-proxyfrom example commands in cli-reference.md.2. Entire Cloud Hypervisor flag family undocumented
src/cli-options.tslines 191–202 define 11--cloud-hypervisor-*flags (binary, kernel, rootfs, supervisor, vcpus, memory-mib, api-timeout-ms, preview, plus 5-sha256digest variants).docs/usage.mdorcli-reference.md. Only a high-level architectural mention exists in CLAUDE.md (linking todocs/cloud-hypervisor-foundation.md).docs/cloud-hypervisor-foundation.mdfrom both.3. Token-steering/budget flags undocumented or partially documented
src/cli-options.tslines ~415–447.--enable-token-steeringis in cli-reference.md but missing from usage.md. Its companion flags--max-cache-misses,--max-model-multiplier,--max-model-multiplier-cap, and--max-permission-deniedare missing from all docs.--enable-token-steering, and add--enable-token-steeringitself to usage.md.4. Vertex AI proxy flags undocumented
src/cli-options.ts(--vertex-api-target,--vertex-api-base-path).--gemini-api-target,--openai-api-target,--anthropic-api-target, etc.) are documented in both usage.md and cli-reference.md, but the Vertex AI variants are absent from all docs.5. Cross-doc drift between usage.md and cli-reference.md
--anthropic-auto-cache,--anthropic-cache-tail-ttl,--docker-host-path-prefix,--network-isolation,--pids-limit,--topology-attach(all in usage.md, missing from cli-reference.md);--enable-token-steering(in cli-reference.md, missing from usage.md).6.
--reflectdocumented only in README.md--reflectappears in README.md but not in the two detailed reference docs (usage.md, cli-reference.md), the reverse of the usual gap pattern.--reflectto usage.md and cli-reference.md with a description matching README's usage.7. Minor undocumented flags
--container-runtime,--no-network-isolation,--anthropic-api-auth-header,--openai-api-auth-headerare absent from all docs.📋 Recommendations
--enable-api-proxydocumentation acrossdocs/usage.md,cli-reference.md, andCLAUDE.md/AGENTS.md— this is actively misleading since it describes opt-in behavior that no longer exists and doesn't warn about the--no-enable-api-proxyhard error.--max-cache-misses,--max-model-multiplier,--max-model-multiplier-cap,--max-permission-denied), and the Vertex AI target/base-path flags.--reflectto the detailed reference docs; document--container-runtime,--no-network-isolation, and the*-api-auth-headerflags.📁 Files Analyzed
src/cli.ts,src/cli-options.ts,src/option-parsers.ts,src/commands/subcommands.ts,src/commands/validators/security-mode.ts(implementation)docs/usage.mddocs-site/src/content/docs/reference/cli-reference.mdREADME.mdAGENTS.md/CLAUDE.mdAll reactions