akiiwa-games/AgentCompass / packaging/verify-release
verify-release
VerifiedPre-release behavioral verification for AgentCompass (router + tray) and usage-float. Installs from the built release artifacts (wheel and Inno Setup installer) into clean, foreign-looking machine profiles and Windows Sandbox, proving user-visible behaviors before tagging or syncing to another computer.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 3.9 KB |
| features.json | config | 10.3 KB |
features.json
View on GitHub{
"schema_version": 1,
"product": "AgentCompass + usage-float",
"entry_points": [
"python -X utf8 packaging/verify_release.py",
"python -X utf8 packaging/foreign_machine_check.py",
"python -X utf8 packaging/zero_dispatch_check.py",
"python -X utf8 packaging/windows_sandbox_check.py --connect"
],
"features": [
{
"id": "F01",
"name": "Stale install & published reference data channel",
"harness": "packaging/foreign_machine_check.py::verify_f01_published_reference_data",
"behaviors": [
"Installed wheel runs from site-packages, never the repo checkout, and reads the published set of its schema from origin.py's PUBLIC_DATA_URL (published/<schema>/ on main)",
"agentcompass data --refresh requests the manifest and only the changed file of that schema's set, with no Authorization header, verifies sha256, accepts the set with the install's own readers and records it under ROUTER_STATE_DIR/public-data/sets/<version>",
"providers, plan --demo --dry-run and doctor request nothing more while no check is due; doctor reports Reference data: fetched <version> (..., schema <shipped>, ...)",
"A tampered sha256 and corrupt JSON fail, and a consistent set no reader can route with is rejected, without replacing the accepted set",
"A newer data schema returns schema_too_new after requesting only the manifest; the accepted set stays",
"A set an older release fetched is kept and doctor names no upgrade item for the channel's state",
"AGENTCOMPASS_PUBLIC_DATA=off makes no request, reads the shipped set and reports - channel off in doctor"
]
},
{
"id": "F02",
"name": "Fleet parity on a clean install without accounts.local.json",
"harness": "packaging/foreign_machine_check.py::verify_f02_fleet_parity",
"behaviors": [
"No accounts.local.json exists in the built wheel or foreign profile",
"Clean discovered accounts route across bounded_coding, deep_debugging, and frontend_visual task archetypes with positive scores"
]
},
{
"id": "F03",
"name": "Classifier cascade across multi-home accounts and families",
"harness": "packaging/foreign_machine_check.py::verify_f03_classifier_cascade",
"behaviors": [
"When JEV_API_KEY is absent and ~/.claude is signed out while ~/.claude-2 is signed in, score_task skips jev and ~/.claude and classifies with claude_haiku_low on account claude-2",
"When all Claude homes are signed out, score_task cascades to codex_terra_low on codex-2 without failing"
]
},
{
"id": "F04",
"name": "Quota, window scope, gate parity, control messages, model management & twin suppression",
"harness": "packaging/foreign_machine_check.py::verify_f04_quota_gate_scope_and_twins",
"behaviors": [
"Pi signed into the same Codex subscription account_id produces no twin bundles alongside codex",
"Antigravity 3p windows on a template-only host have positive capacity priced from the pool prior",
"On Antigravity with 3p-weekly used up, a Gemini launch is allowed and a Claude-model launch is blocked by both routing and router_gate_hook",
"A 5h window at its floor blocks a new launch for claude, codex, and claude-2 (100% blocks all; 75% blocks claude/claude-2 Pro 25% floor)",
"Active in-flight ledger reservations count against quota in router_gate_hook and secondary Claude accounts not named claude-2 resolve from bundle/labels",
"A Magpie pool with accounts: 2 is judged on N*(100-mean)",
"A new data-declared scoped window works with zero code change",
"The owning parent's control message to its running worker on an exhausted window passes, while messages to idle, unknown-status, and non-owner workers are blocked",
"Superseded discovered models default to unchecked, latest & unknown new models default to checked, user checkbox overrides both ways, tray agrees with routing, and checked metered models are chosen only when no subscription is eligible"
]
},
{
"id": "F05",
"name": "Antigravity plan tier from Google loadCodeAssist",
"harness": "packaging/foreign_machine_check.py::verify_f05_antigravity_plan_tier",
"behaviors": [
"antigravity_plan resolves exact tier ('pro' / 'ultra5' / 'ultra20') from loadCodeAssist paidTier.id using agy's keyring token",
"antigravity-plan.json is cached under ROUTER_STATE_DIR without leaking the OAuth token",
"AGENTCOMPASS_ANTIGRAVITY_PLAN=off disables the Google tier lookup"
]
},
{
"id": "F06",
"name": "Magpie gateway pools & stale reading POST refresh",
"harness": "packaging/foreign_machine_check.py::verify_f06_magpie_pools_and_refresh",
"behaviors": [
"Auto-discovered Magpie gateway calls POST /v1/magpie/quotas/refresh for hours-old readings on force=True",
"Largest plan wins and shows pool multiplier 'pro ×2' with window labels '7d×2' and '5h×2'",
"usage_details.members masks account emails (a***@...) and never stores raw emails or tokens"
]
},
{
"id": "F07",
"name": "Stranger account topologies & setup/discovery/uninstall home parity",
"harness": "packaging/foreign_machine_check.py::verify_f07_stranger_topologies",
"behaviors": [
"Across only-Claude, only-Codex, Magpie-Claude, agy-not-signed-in, no-Paseo, and Paseo custom CLAUDE_CONFIG_DIR + 2-hop extends chain profiles, setup writes rules/skills/hooks to the exact set of homes provider_inventory.discover() finds",
"Each Claude/Codex home keeps its own settings (model, effortLevel, env) when setup adds the gate hook, and uninstall restores them unchanged",
"agentcompass uninstall --yes removes every setup-created file and restores pre-existing files unchanged",
"Re-running setup --yes --hooks after a previously managed custom home is retired cleans up its managed doc/hook/skill link and drops it from setup-manifest.json",
"only-Claude and only-Codex profiles classify and route cleanly without Paseo or API keys"
]
},
{
"id": "F08",
"name": "Tray account cards & hover + usage-float standalone and hosted rows",
"harness": "packaging/foreign_machine_check.py::verify_f08_tray_cards_and_usage_float",
"behaviors": [
"agentcompass setup --usage-float clones usage-float into ~/.agent-compass/plugins/usage-float",
"Tray account_cards() shows remaining_pct, 'pro ×2', '7d×2'/'5h×2' tags, no window tag on week-only accounts, and masked member hover tooltip",
"Hosted usage-float shares ~/.agent-compass/usage-cache.json, displays left_pct=150%/100% on ×2 pools, titles week-only accounts without a '7d' suffix, and renders masked member hover detail",
"usage_float.parse_claude_usage sets models=['sonnet'] on weekly_sonnet and window_scope.binds scopes it to Sonnet only"
]
},
{
"id": "F09",
"name": "Real machine isolation & process hygiene",
"harness": "packaging/foreign_machine_check.py::main",
"behaviors": [
"Real user profile files (~/.agent-compass, ~/.claude*, ~/.codex*, ~/.paseo) are unpolluted",
"No background process started inside the foreign-machine sandbox remains running"
]
},
{
"id": "F10",
"name": "Evidence-calibrated routing after 0.3.6",
"harness": "packaging/foreign_machine_check.py::verify_f10_evidence_calibrated_routing (probe: packaging/f10_evidence_probe.py)",
"behaviors": [
"On a first scan a model public data lists in `superseded` (an older Haiku beside Haiku 5.5) is discovered unchecked and its successor checked",
"Metered accounts are a last resort (excluded while a subscription can take the work, chosen only when none can) and never-billed (unknown billing) accounts are explicit selection only",
"Every check above is run once more against data broken inside the throwaway profile and must fail there"
]
},
{
"id": "F11",
"name": "Oh My Pi OAuth subscription logins",
"harness": "packaging/foreign_machine_check.py::verify_f11_omp_oauth_subscriptions (probe: packaging/f11_omp_probe.py)",
"behaviors": [
"An enabled omp OAuth login (agent.db auth_credentials) signed in as the native Codex account is a subscription account whose quota_account is that account, and discovery output carries no omp token or email",
"Its bundles sit on the Codex account with omp's `<provider>/<model>` launch ids and no twin of a (harness, model, effort)",
"The plan treats an omp subscription bundle like Pi's: eligible, drawing the shared Codex window, excluded when that window is empty",
"With only subscription logins (no models.yml, no stored API key) there is no metered `omp` account in discovery or routing; a stored API key keeps it beside the login",
"An omp-only login is its own subscription account on the Codex plan's windows with omp bundles",
"A disabled credential is ignored; two logins of one provider are two accounts; a corrupt or locked agent.db is missing_credentials, not a crash",
"Every check above is run once more against broken input inside the throwaway profile and must fail there"
]
},
{
"id": "F12",
"name": "Tray version display, one-click self-update & single-flight tray restart",
"harness": "packaging/foreign_machine_check.py::verify_f12_self_update_and_version_display",
"behaviors": [
"Tray and settings page display AgentCompass X.Y.Z and flag version mismatch when tray and router (skills, gate hooks, or PATH) differ",
"Release check caches GitHub Releases latest for 6h under STATE_DIR and shows '可更新到 vX.Y.Z' when a newer release exists",
"One-click Update AgentCompass detects install shape (installer, wheel, git), refuses missing or mismatched sha256 digests, and upgrades using the owning tool (uv, pip, git)",
"An installed older wheel (0.3.7) in a foreign profile upgrades to the locally built newer wheel (0.3.8) through update_agentcompass, clears the version mismatch, and runs the post-update step (agentcompass data --refresh), which checks the published reference data and deletes none of the channel's state",
"Clicking Restart tray is single-flight and emits immediate and post-restart tray notifications"
]
}
]
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |