prizmkit 1.1.166 → 1.1.167

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -1,170 +1,136 @@
1
- # Deployment Policy and Execution Contract
1
+ # Deployment Policy and Execution
2
2
 
3
- Use this contract for every deploy or mutating online operation, regardless of whether its capabilities are verified, dynamic, manual, or mixed. It also determines when an apparently read-only request must become a mutating plan.
3
+ Use this reference for a deployment write, material producer, migration, redeploy, promotion, dependency change, decommission, or optional automation setup. The selected action comes from the capability contract; this reference owns execution and conditional Gate policy.
4
4
 
5
- ## Side-Effect Classification
5
+ ## Invocation Policy
6
6
 
7
- | Class | Examples by effect | Policy |
8
- |---|---|---|
9
- | `read_only` | Project-native file reads, non-mutating capability/version/auth identity probes, status, configured health reads, bounded logs, history, validation, bounded diagnostics | May run directly when proven project-scoped and read-only. Do not update the local record as a side effect. |
10
- | `reversible_write` | Create a release with verified cleanup, start/restart a project unit, update sole local record, apply project-scoped target configuration with verified restore | Show one complete immutable plan and obtain explicit confirmation before any step. |
11
- | `high_impact` | Project/source file change, privileged/global or compatibility-sensitive dependency change, shared infrastructure, traffic/data-affecting change, plan expansion, unknown-outcome replay, no reliable rollback | Obtain the complete-plan confirmation plus a dedicated confirmation at the exact boundary. |
12
- | `destructive` | Irreversible migration, data/resource deletion, overwrite without verified restore, destructive rollback | Obtain a dedicated confirmation identifying exact data/resources and recovery; refuse when impact cannot be bounded. |
7
+ An explicit deploy invocation authorizes the selected Profile's standard deployment envelope. Show a compact execution summary for awareness, then continue without asking the user to approve the same ordinary plan again.
13
8
 
14
- A command name, “dry-run” flag, target label, or recipe classification does not prove read-only behavior. Inspect documented effects and actual tool capability.
9
+ `deploy plan` is read-only. It reports the selected Profile, observed current state, latest allowed desired release, expected actions, downtime/data/rollback/resource/cleanup effects, verification, and optional automation opportunity without performing writes.
15
10
 
16
- ## Immutable Plan
11
+ The standard envelope covers project-owned, bounded actions needed to build/package, resolve project-declared dependencies, apply non-destructive native migrations, create a candidate, start/restart, hand off live traffic, verify, retain bounded rollback, retire exact old runtime resources, clean failed/temporary resources, and update the selected Profile record.
17
12
 
18
- The complete preview has one `plan_id` and a Secret-free digest over:
13
+ Pause with one exact `WAITING_USER` action only for:
19
14
 
20
- - current project root and project/revision evidence;
21
- - target profile, endpoint/connection identity, environment, and actual observed ports;
22
- - requested operation and release/artifact identity;
23
- - all capability descriptors and ordered steps;
24
- - project/user dependencies and installed versions;
25
- - database/data analysis;
26
- - health boundary and verification layers;
27
- - rollback reference or explicit no-rollback state;
28
- - retry ceilings, human checkpoints, cleanup, and residual-resource policy;
29
- - local-record and optional project-file effects.
15
+ 1. an identity or authentication action that only the user can complete;
16
+ 2. a target choice current evidence cannot resolve;
17
+ 3. an irreversible data effect or data impact that cannot be bounded safely;
18
+ 4. a shared or unknown-ownership resource decision.
30
19
 
31
- Do not include Secret values in the digest input or preview. Bind only Secret names and external source references.
20
+ A missing reliable rollback is disclosed but does not create another approval. Privileged/global work not already established by the selected target boundary is treated as an authorization or shared-ownership decision rather than silently executed.
32
21
 
33
- Every ordered step contains every field below:
22
+ ## Effect and Resource Policy
34
23
 
35
- | Field | Required content |
24
+ Read-only project-scoped observations may run directly when their selected command is known, output is bounded, and no durable cache, temporary resource, material traffic, or charge is created.
25
+
26
+ Resource impact has three values:
27
+
28
+ | Impact | Policy |
36
29
  |---|---|
37
- | `capability` | Semantic capability name. |
38
- | `availability` | `verified`, `dynamic`, `manual`, or `unsupported`. |
39
- | `evidence` | Bounded origin/observation/method/freshness facts supporting this exact step. |
40
- | `preconditions` | Rechecked immediately before execution. |
41
- | `inputs` | Non-sensitive values and Secret references, with source and identity. |
42
- | `effects` | Files, runtime units, releases, data, traffic, network, credential, cost, and residual resources affected. |
43
- | `risk` | `read_only`, `reversible_write`, `high_impact`, or `destructive`, plus reason. |
44
- | `confirmation` | `none`, `complete_plan`, and/or exact dedicated reason(s). |
45
- | `execution` | Complete executable/arguments, target-native action, or human action; target and working directory are explicit. |
46
- | `verification` | Required and optional observable postconditions. |
47
- | `failure_handling` | Stop, cleanup, inspection, rollback offer, and manual recovery behavior. |
48
- | `idempotency` | `proven`, `not_idempotent`, or `unknown`, with evidence. |
49
- | `retry` | Maximum attempts, retryable conditions, backoff/stop condition, and no-retry conditions. |
50
- | `unknown_outcome_inspection` | Proven read-only observation that distinguishes applied/not-applied/indeterminate before replay. |
30
+ | `negligible` | Small bounded observation with no durable material output; execute directly. |
31
+ | `material` | Finite build, test, cache, transfer, provider, or temporary-resource work; execute inside one internal aggregate budget without another confirmation. |
32
+ | `unbounded` | A material dimension, repetition, ownership, cleanup, or stop boundary is unknown; reduce to a finite action or request one human decision. |
51
33
 
52
- An unsupported required step makes the plan non-executable. A manual step is a human checkpoint and does not disappear from the order.
34
+ A material budget covers aggregate attempts rather than resetting per retry. Establish enough controller/target capacity, timeout/attempt limits, owned temporary outputs, cleanup, and stop thresholds to prevent uncontrolled work. Do not free capacity by pruning unowned, global, persistent, or shared resources.
53
35
 
54
- Confirmation applies only to the exact digest. A changed command, argument, target identity, project revision, release input, effect, risk, dependency, migration, verification boundary, rollback state, or added step creates a new plan and invalidates prior confirmation.
36
+ Do not run a material producer when current project-native evidence already covers the same source/release, toolchain, target assumptions, and required scope. Run the least expensive missing producer at most once for that binding. A changed binding starts a newly observed current→desired decision; it does not justify automatic repeated full rings.
55
37
 
56
- ## Confirmation Matrix
38
+ ## Current → Desired Loop
57
39
 
58
- | Trigger | Confirmation required | Evidence shown before confirmation |
59
- |---|---|---|
60
- | Proven read-only and project-scoped | `none` | Target and bounded read effects |
61
- | Ordinary reversible writes | `complete_plan` | Entire ordered plan, effects, verification, rollback/cleanup, and digest |
62
- | Create or modify `prizmkit.deploy.json` or another project/source file | `PROJECT_FILE_CHANGE` | Exact complete file preview/diff, path, ownership, and rollback |
63
- | Privileged/global or compatibility-sensitive install/upgrade | `PRIVILEGED_OR_GLOBAL_DEPENDENCY` | Exact dependency/version/source, privilege, compatibility, target scope, effects, and uninstall/restore path |
64
- | Destructive or irreversible data migration | `DESTRUCTIVE_OR_IRREVERSIBLE_MIGRATION` | Exact migration, affected data, reversibility, locks/downtime, backup/restore evidence, and stop point |
65
- | Shared infrastructure or resource | `SHARED_INFRASTRUCTURE` | Exact shared resource, owners/consumers, traffic/cost/blast radius, and restore plan |
66
- | New scope or changed immutable plan | `PLAN_EXPANSION` | Complete replacement plan and change summary; old confirmation is void |
67
- | Possible prior write with unknown outcome | `UNKNOWN_WRITE_REPLAY` | Read-only inspection result, duplicate/replay effects, and exact replay action |
68
- | Deployment without verified rollback | `NO_RELIABLE_ROLLBACK` | Live-state risk, failure effects, residual resources, and manual recovery procedure |
40
+ For one selected Profile:
69
41
 
70
- A dedicated confirmation supplements, not replaces, complete-plan confirmation. Capture the user's actual affirmative response and exact plan/reason scope in the active interaction. For interruption recovery, persist only a non-sensitive receipt containing confirmation kind, matching plan digest, and confirmation time; never persist unrelated conversation or Secret values. A request to deploy, initial feature request, silence, elapsed time, earlier approval, or approval of a different digest is not confirmation.
42
+ 1. Observe authoritative target state, project revision, automation, live selector, owned resources, persistent/shared dependencies, and existing reusable evidence.
43
+ 2. Bind the Profile policy's latest eligible desired release and keep it fixed for this operation.
44
+ 3. If the same release is already live and required minimum health passes, perform no material build/package/release/restart action.
45
+ 4. Compute the remaining difference and choose the smallest proven next action.
46
+ 5. Activate only the Resource, Supersession, Data, Shared Resource, Human Checkpoint, or Unknown Outcome Gate that applies.
47
+ 6. Execute once within its aggregate budget, then observe before selecting another action.
48
+ 7. Continue until desired is live and the Profile inventory converges, or until one human action/no safe continuation remains.
71
49
 
72
- If confirmation is unavailable in a non-interactive run, stop before the action with `WAITING_USER`, persist only confirmed minimal recovery context, and state the exact decision needed.
50
+ When an active recoverable operation already targets the same Profile and desired release, attach to or resume that operation instead of creating another candidate. Different Profiles may progress independently; do not use one Profile's current/recovery as another Profile's state.
73
51
 
74
- ## Dependencies and Installation
52
+ The target control plane and actual workload state outrank the Profile record. Reconcile drift before writing. A target-fingerprint mismatch or unexplained ownership conflict requires one target/ownership decision.
75
53
 
76
- 1. Read project-native dependency/runtime declarations and lock/version files.
77
- 2. Distinguish project-local, user-local, target-local, global, and privileged dependencies.
78
- 3. Inspect actual installed versions and capabilities read-only.
79
- 4. Prefer the project's declared package/dependency mechanism and pinned compatibility.
80
- 5. Include every install, upgrade, source, version, scope, write, service restart, and restore path in the plan.
81
- 6. Obtain `PRIVILEGED_OR_GLOBAL_DEPENDENCY` confirmation for privileged/global or compatibility-sensitive changes.
82
- 7. Never silently install, upgrade, replace, remove, or select a guessed/wildcard version.
83
- 8. Never transfer a controller-built dependency directory across systems unless project-native evidence explicitly establishes artifact portability and the confirmed package contract includes compatibility/integrity verification. Prefer reproducible target-compatible artifacts or target-native dependency resolution.
54
+ ## Supersession Gate
84
55
 
85
- A missing tool can make a capability `manual` or `unsupported`; it is not permission to install it.
56
+ When a current deployment exists, inventory exact live, candidate, rollback, superseded, failed, temporary, and orphaned project-owned units plus the live selector. Separate persistent data and shared dependencies before retirement.
86
57
 
87
- ## Secret, Log, and Diagnostic Policy
58
+ Prefer an evidence-supported candidate/handoff transition:
88
59
 
89
- - Secret values remain in user/target-owned Secret systems. Plans contain names and source references only.
90
- - Never print or persist credentials in URLs, authorization headers, tokens, keys, cookies, connection strings, environment assignments, command arguments, or raw diagnostic output.
91
- - Before display, redact recognized credential syntax and user-identified sensitive patterns. Omit unnecessary lines rather than retaining masked copies.
92
- - Logs and diagnostics specify project-owned source, purpose, time range and/or line limit, filters, and redaction. Default to the smallest useful bounded window; do not stream indefinitely.
93
- - Do not run broad process, filesystem, network, account, or infrastructure inventory when a project-scoped observation answers the question.
94
- - If redaction cannot be made reliable, do not display or persist the content; provide a manual inspection checkpoint.
60
+ 1. keep the proven old live release;
61
+ 2. create desired candidate without disturbing it;
62
+ 3. run candidate pre-handoff checks;
63
+ 4. switch the supported live/traffic selector;
64
+ 5. verify desired through the actual live path;
65
+ 6. retain the Profile's bounded rollback set; when neither project-native nor Profile policy defines one, retain one verified previous release if the target supports it, otherwise disclose rollback absence and continue without another confirmation;
66
+ 7. retire exact superseded, failed, temporary, and orphaned runtime resources;
67
+ 8. observe the converged inventory.
95
68
 
96
- ## Database and Persistent-Data Gate
69
+ Use in-place replacement only when the target lacks coexistence and its downtime/data effects are known. Missing rollback alone does not pause, but irreversible data or unknown/shared effects do.
97
70
 
98
- Before any migration or data-affecting release step, report:
71
+ Candidate failure before handoff preserves old live and cleans exact candidate resources. A post-handoff ordinary failure follows forward progress for the same desired release. Emergency isolation or a temporary traffic return is allowed only to contain data, security, or ongoing user harm; it is not terminal deployment success.
99
72
 
100
- 1. exact migration identity and target data store;
101
- 2. forward and reverse operations, or why reverse is impossible;
102
- 3. tables/collections/files/queues and estimated data transformation/loss impact;
103
- 4. lock level, downtime/online behavior, duration uncertainty, and traffic sequencing;
104
- 5. compatibility window between old/new code and old/new schema;
105
- 6. backup identity, completion, restore procedure, and restoration verification;
106
- 7. expand/contract or other safer alternative when available;
107
- 8. stop conditions, unknown-outcome inspection, and recovery owner.
73
+ Persistent data, shared infrastructure, and unproven resources are never classified as old application runtime merely because code changed.
108
74
 
109
- A migration tool reporting success does not prove data correctness. Verify schema/state and application health. Destructive or irreversible migration requires `DESTRUCTIVE_OR_IRREVERSIBLE_MIGRATION` confirmation. Never suppress a target/tool's own safety prompt.
75
+ ## Data and Dependency Gates
110
76
 
111
- ## Network and Shared Infrastructure
77
+ Use project-native locked dependency mechanisms and actual installed capability. Project-local or target-local declared resolution within the selected deployment is standard. Never silently install or replace a privileged/global dependency, guess a wildcard version, or transfer a controller dependency directory whose target compatibility is unproven.
112
78
 
113
- Use actual observed connection and service ports. Distinguish controller-side connection endpoints from target-side application/listener endpoints. Do not infer ports from a platform label or fixed recipe.
79
+ For a data-affecting action, establish migration identity, old/new compatibility, locks/downtime, irreversible transformations, backup/restore evidence when required, traffic sequencing, post-migration verification, and unknown-outcome inspection. Non-destructive native migration can remain in the standard envelope. Irreversible or insufficiently bounded impact pauses for one data decision.
114
80
 
115
- Any routing, certificate, firewall, load-balancer, domain, shared network, shared storage, shared database, or shared runtime change lists affected consumers and requires `SHARED_INFRASTRUCTURE` confirmation when it can affect another workload. General hardening or unrelated server administration is out of scope.
81
+ Do not treat a routing, certificate, firewall, load-balancer, domain, shared network/storage/database, or shared runtime change as project-owned until current evidence proves ownership and consumers. Shared or unknown ownership pauses before mutation.
116
82
 
117
- ## Operation Parity
83
+ ## Build Once, Promote Many
118
84
 
119
- Every intent uses the same capability, planning, confirmation, and verification contracts:
85
+ When the project produces an immutable artifact, build it once for one source/toolchain binding. Test, staging, production, and custom Profiles should promote the same verified artifact identity when their release policies and target compatibility allow it. Do not rebuild merely because the destination Profile changed.
120
86
 
121
- | Operation | Minimum behavior |
122
- |---|---|
123
- | `deploy` | Discover/validate; compile all build/package/transfer/configure/migrate/release/start/health dependencies that evidence requires. |
124
- | `status` | Observe current release/runtime state read-only; report inaccessible layers. |
125
- | `health` | Execute configured health boundary read-only and preserve required/optional distinction. |
126
- | `logs` | Read a bounded redacted project-owned window. |
127
- | `stop` | Observe the current project-owned unit, confirm the stop effects and rollback/start recovery, stop only that unit, and verify the intended offline/live state. |
128
- | `restart` | Observe current unit, confirm reversible write, restart only that unit, verify runtime/health/log layers. |
129
- | `rollback` | Verify rollback reference and current live state, confirm effects, restore, then run full applicable verification. |
130
- | `history` | Read concise bounded local and target-native release evidence without writing event files. |
131
- | `validate` | Check evidence, configuration, capability prerequisites, policy, and rollback without deployment writes. |
132
- | `diagnose` | Gather bounded read-only evidence; compile a new plan before any repair. |
133
- | `takeover` | Discover ownership/conflicts/backups/coexistence; dedicated confirmation applies to shared or non-owned changes. |
134
- | `resume` | Rediscover and compare evidence/digest before continuing; never assume the prior process completed nothing. |
135
- | `cancel` | Stop future actions and perform only already-confirmed safe cleanup; report residual resources. |
136
-
137
- ## Execution Algorithm
138
-
139
- 1. Revalidate project, target, authentication identity, current live state, and the step's evidence/preconditions.
140
- 2. Verify that complete-plan and applicable dedicated confirmations match the current digest.
141
- 3. Report `step N/total`, capability, action class, target boundary, and evidence being sought. Do not promise a universal duration.
142
- 4. Execute exactly the planned action. Do not improvise flags, repairs, cleanup, scope, or dependencies.
143
- 5. Capture only bounded redacted evidence and determine the immediate command/target result.
144
- 6. Run the step verification before advancing.
145
- 7. Atomically update the sole local record only at confirmed plan points; validate before and after write.
146
- 8. Stop at a human checkpoint, failure, unknown outcome, invalidated plan, cancellation, or terminal verification.
147
-
148
- ## Retry and Unknown Outcomes
149
-
150
- - Automatic retry is permitted only for a proven read-only action or a proven idempotent action under its explicit maximum-attempt and stop-condition policy.
151
- - Retry count is bounded in the immutable plan; never convert a bounded retry into an open loop.
152
- - Authentication denial, invalid input, destructive failure, plan mismatch, failed required verification, and unknown write outcome are not automatic-retry conditions.
153
- - After timeout, disconnect, or lost response during a possible write, run the planned read-only inspection. Classify the effect as applied, not applied, or indeterminate.
154
- - `applied`: verify postconditions; do not replay.
155
- - `not applied`: retry only if idempotency/retry policy allows and confirmation remains valid.
156
- - `indeterminate`: stop with `BLOCKED` or `WAITING_USER`; any replay requires `UNKNOWN_WRITE_REPLAY` confirmation.
157
-
158
- ## Deterministic Policy Cases
159
-
160
- | Case | Required outcome |
161
- |---|---|
162
- | Read-only project status with proven non-mutating probe | Execute directly; no local-record write |
163
- | Ordinary reversible restart | Complete-plan confirmation before write |
164
- | User requested deploy but has not approved compiled plan | `WAITING_USER` |
165
- | Plan gains an unpreviewed configuration write | New plan plus `PLAN_EXPANSION`; no execution |
166
- | Required privileged dependency is missing | Dedicated confirmation or `WAITING_USER`; no silent install |
167
- | Migration is irreversible and impact cannot be bounded | `BLOCKED` |
168
- | Prior write timed out and inspection is indeterminate | No replay; `BLOCKED` or `WAITING_USER` |
169
- | Rollback is absent but exact risk is accepted | Deployment may proceed; retain explicit manual recovery |
170
- | Required health check fails after release command succeeds | `FAILED`; command success is not operation success |
87
+ A production-like Profile may require an artifact already verified by another Profile. Promotion still observes the destination current state and performs only its missing transfer/release/configure/handoff/verification work.
88
+
89
+ When no artifact mechanism exists, use the current committed revision and the minimum project-native producer. Do not silently include dirty workspace content.
90
+
91
+ ## Retry, Unknown Outcomes, and Forward Work
92
+
93
+ Retry only a proven read-only or idempotent action for a transient condition within its aggregate attempt/resource budget. After attempts are exhausted, select a bounded diagnostic or a different proven forward action; do not reset the same wrapper into an open loop.
94
+
95
+ After timeout, disconnect, or lost response during a possible write, inspect current target state before replay:
96
+
97
+ - applied: verify and continue without replay;
98
+ - not applied: retry only when the action remains idempotent and within budget;
99
+ - indeterminate: return one `WAITING_USER` unknown-outcome action rather than duplicating state.
100
+
101
+ Authentication denial, invalid input, target mismatch, destructive uncertainty, and unknown ownership are not automatic retry conditions.
102
+
103
+ ## Secret, Log, and Output Policy
104
+
105
+ Secret values stay in external user/target systems. Actions and managed records contain only Secret names, source references, and presence. Omit credentials from URLs, headers, arguments, environment output, logs, errors, and summaries.
106
+
107
+ Bound logs by source, purpose, time and/or line count. Prefer omission when reliable redaction is unavailable. Do not run broad account, process, filesystem, network, or infrastructure inventory when a project-scoped observation answers the current question.
108
+
109
+ ## Post-Success Automation
110
+
111
+ Offer automation only after deployment has already reached `SUCCEEDED` and only when project-native evidence shows no suitable automation exists. The offer is a separate optional project-file change, not a condition of deployment success.
112
+
113
+ ### Remote Profile
114
+
115
+ When repository evidence identifies GitHub and the target supports non-interactive deployment, offer GitHub Actions. Ask the user whether to configure it and derive the trigger from project release policy; if no policy resolves main-branch, tag/release, or manual dispatch behavior, ask one trigger question.
116
+
117
+ Separate quick push and pull-request checks from CD. CD should reuse a verified immutable artifact when available and perform only the selected Profile's missing deployment, verification, and convergence work. Keep credentials in repository Environments/Secrets or the target Secret system. A production approval, when requested by the user, belongs to repository environment policy rather than repeated Deploy Skill confirmation.
118
+
119
+ For another repository host, offer only a semantically equivalent automation proven by current project and host capability. Do not route by provider name alone.
120
+
121
+ ### Local Profile
122
+
123
+ Do not create remote CI/CD for local deployment. Offer the best existing project-native local loop in this order when supported: watch or hot reload, incremental build and restart, container file sync/watch, then a lightweight commit-triggered local deploy.
124
+
125
+ Local automation affects only the selected local Profile. It does not run production migration, access a remote Profile, or execute a complete release-candidate/production ring on every file save. When no incremental capability exists, keep manual one-click deploy rather than inventing scripts.
126
+
127
+ Do not create a PrizmKit automation-state artifact. Detect automation from project-native workflow/config files. A declined or failed setup does not change the completed deployment result.
128
+
129
+ ## Profile Lifecycle
130
+
131
+ Removing a declaration entry or local Profile record does not remove a target deployment.
132
+
133
+ - **Decommission:** inventory exact Profile-owned runtime resources, obtain any data/shared decision that applies, remove only authorized target resources, and verify the intended offline state.
134
+ - **Forget:** remove only local PrizmKit management evidence after disclosing that the target is untouched.
135
+
136
+ Never infer decommission from file deletion or forget from a failed target cleanup.
@@ -2,14 +2,14 @@
2
2
 
3
3
  Local fixtures and side-effect-free contract tests may run without a deployment confirmation. A test crosses the real-resource boundary if it connects to or authenticates against a server, remote target/control plane, real database, credential store, external URL that creates meaningful traffic, or any resource that may create state, traffic, side effects, quota use, or cost.
4
4
 
5
- A “smoke test,” read-only label, sandbox label, free tier, dry-run flag, or non-production name does not bypass this boundary.
5
+ A “smoke test,” read-only label, sandbox label, free tier, dry-run flag, or non-production name does not bypass this boundary. An ordinary deploy to a selected test/staging Profile remains a deployment, not a real-resource framework test; environment names never change the route.
6
6
 
7
7
  ## Deterministic Local Tests
8
8
 
9
9
  These may run normally when they use only local temporary fixtures and no external credentials or network:
10
10
 
11
11
  - capability availability and evidence-precedence cases;
12
- - immutable plan and confirmation-policy cases;
12
+ - current→desired convergence and conditional-Gate cases;
13
13
  - operation/command/config schema parity;
14
14
  - JSON schema closed-shape and terminal-status checks;
15
15
  - Secret-like fixture redaction and persistence rejection;
@@ -38,18 +38,18 @@ Before any real-resource test, present one dedicated preview containing every fi
38
38
 
39
39
  Ask for separate explicit authorization of this exact preview. Approval of a feature, deployment plan, local test suite, or earlier smoke test is not real-resource test authorization. Any changed target, identity, action, effect, duration, cleanup, cost, or scope invalidates authorization.
40
40
 
41
- In a non-interactive session without exact authorization, do not connect. Return `WAITING_USER` with the non-sensitive preview or `BLOCKED` when the target/effects cannot be bounded.
41
+ In a non-interactive session without exact authorization, do not connect. Return `WAITING_USER` with the non-sensitive preview when one decision can authorize the test, or `FAILED` when no safe bounded test path exists.
42
42
 
43
43
  ## Execution
44
44
 
45
45
  When authorized:
46
46
 
47
47
  1. revalidate target and authentication identity read-only;
48
- 2. apply the ordinary deployment policy and all dedicated confirmations in addition to this test authorization;
48
+ 2. apply the ordinary deployment policy and every conditional Gate implicated by the test in addition to this separate authorization;
49
49
  3. execute only the previewed bounded actions;
50
50
  4. report progress without a universal duration promise;
51
51
  5. stop on any previewed condition, plan mismatch, unknown write outcome, credential boundary change, or unexpected cost/traffic/state;
52
- 6. run layered verification and bounded redaction;
52
+ 6. run the target-appropriate required verification observations and bounded redaction;
53
53
  7. perform only authorized cleanup;
54
54
  8. verify cleanup independently and report residual resources.
55
55
 
@@ -66,7 +66,7 @@ Report the test result separately from cleanup:
66
66
  - cleanup actions and verification;
67
67
  - every residual resource and manual cleanup step.
68
68
 
69
- If the test passes but cleanup fails or cannot be verified, do not report an unqualified success. Return `FAILED` for a required cleanup failure or `BLOCKED` for an inaccessible/indeterminate cleanup boundary, and state that resources may remain.
69
+ If the test passes but cleanup fails or cannot be verified, do not report an unqualified success. Return `WAITING_USER` when one exact inspection or cleanup action can resolve the residual, otherwise return `FAILED`; state that resources may remain.
70
70
 
71
71
  ## This Framework Feature's Test Boundary
72
72