vinyasar 0.3.0__tar.gz

vinyasar-0.3.0/CHANGELOG.md

@@ -0,0 +1,126 @@
+# Changelog
+
+## 0.2.2
+
+Consolidated native macOS Darwin arm64 adversarial qualification suite (18 attack
+vectors covering path symlink boundaries, multi-process races, reservation crash
+semantics, staged publication healing, digest tampering, payload fuzzing, and MCP
+argument containment). Total Vinyasar test count expanded to 123 tests (257 total tests
+with Yasarda 0.3.1). Confirmed clean Darwin arm64 platform qualification with zero runtime
+regressions. No changes to Yasarda 0.3.1 or transaction/profile/run schemas.
+
+## 0.2.1
+
+Canonicalize host-supplied repository, state, report and environment/default
+paths during init only; canonicalize the output parent without following its
+leaf or overwriting existing files. Existing JSON/YAML profiles and their source
+paths remain link-strict. Add read-only SDK/CLI path diagnostics. Bound profile
+reads and reject nonregular files before reading. Add alias, link-loop, race and
+no-write tests. No changes to Yasarda 0.3.1 or transaction/profile/run schemas.
+
+## 0.2.0
+
+Review-driven hardening: exact receipt plus fresh complete verification; host
+recovery file/byte limits; cooperative deadlines and bounded locks; whole-target
+serialization; durable run/action identity; conservative crash reconstruction;
+stable retry budgets and explicit host-only legacy adoption; bounded/redacted
+diagnostics; schema-v2 canonical integrity and references; staged no-overwrite
+report publication; parent-link/read guards; accurate MCP error/replay behavior;
+correct CloudEvents payload schema handling. No new mutation capability.
+
+## 0.1.0
+
+Original bounded supervisor over Yasarda 0.3.0. Original test evidence is retained
+under docs/baseline and is not evidence for this release.
+
+
+# Historical Yasarda Changelog
+
+
+# Changelog
+
+## 0.3.1
+
+Preflight file count, per-file byte size and total bytes on each snapshot side
+before recovery or explicit rollback can restore any file. Existing exact-byte,
+mode, path, policy, journal and divergence guards are unchanged. The plan
+compatibility marker remains 0.2.1; original 0.2.0 journals remain incompatible.
+
+# Changelog
+
+## 0.3.0 - 2026-09-09
+
+Public supervisor boundary for the separately packaged Vinyasar 0.1.0 release.
+
+- Add `Engine.pending_operations()` as the supported read-only recovery-status
+  API; callers no longer need to inspect transaction journals directly.
+- Add `Repository.mutation_filesystem_issue()` so supervisors can explain a
+  storage refusal without attempting mutation.
+- Add `Service.operations()` and advertise `inspect_operations` in capability
+  metadata.
+- Apply the same known cloud/remote filesystem refusal to Yasarda's journal state
+  directory; local repository bytes are not mutated when durable recovery state
+  is configured on unsupported storage.
+- Keep all transaction serialization, staging, rollback, and conditional restore
+  behavior private and unchanged.
+
+## 0.2.1 - 2026-09-09
+
+Patch hardening release; local artifacts only.
+
+- Refuse mutation and recovery in known iCloud/macOS cloud-storage paths and
+  known Linux remote/FUSE mounts unless the host explicitly opts in.
+- Reject a `build_server(repo=Repository(...), policy=...)` configuration when
+  the supplied policy conflicts with the policy already bound to the repository.
+- Deprecate the legacy `allow_commands=True` SDK argument visibly; it still never
+  enables command execution.
+- Add Python 3.14 metadata and regression tests for filesystem detection, host
+  override behavior, MCP policy conflicts, and the deprecated command flag.
+
+## 0.2.0 - 2026-09-09
+
+Hardening of the supplied 0.1.0 package. Local artifacts only; not published.
+
+### Correctness
+
+- Fixed the eight supplied audit cases: repeated replacement, conflicting edits,
+  missing desired-state verification, direct SDK dependency bypass, invalid
+  counts, contradictory hashes, newline rewriting, and symlink-target deletion.
+- Enforced the packaged JSON Schema, duplicate-key checks, strict field types,
+  path validation, graph validation, output-size bounds, and final-state checks.
+- Kept transforms pure; only the shared execution engine writes destination files.
+
+### Execution
+
+- Added immutable saved plans and exact-digest approval, whole-file input/output
+  fingerprints, branch/HEAD binding, scoped dependency snapshots, and mode guards.
+- Added external private state, cooperative writer locks, output staging, journals,
+  verified retries, guarded rollback, and interrupted-operation recovery.
+- Refuses unknown third-party changes, pending recovery, protected paths, bare Git
+  object stores, links, case-fold collisions, and unsupported file types.
+
+### Interfaces
+
+- SDK, CLI, and MCP share Engine and Service; no interface-specific transforms.
+- CLI supports saved-plan execution, receipts, recovery, rollback, and operating
+  modes. Immediate local `apply --yes` remains for trusted callers.
+- MCP is now built into the base distribution as a small stdio tools/resources
+  adapter. No external `mcp` package is needed. Root scope and approval are set by
+  the host; tool callers cannot widen them. MCP defaults to read-only.
+
+### Deliberate compatibility changes
+
+- Multiple transformations of one file are rejected, not sequentially guessed.
+- Unrecognized manifest fields/types fail rather than being ignored/coerced.
+- Arbitrary command verification is unsupported, including when the old SDK
+  `allow_commands=True` argument is passed. Such checks fail before publication.
+- The old MCP `apply_remediation(execute=True, repo_path=..., allow_commands=...)`
+  surface is replaced by saved-plan execution with independent host authorization.
+- `init --force` and implicit catalog overwrites are removed. Author catalogs as
+  explicit files or copy the example into a disposable directory.
+- Mutation is currently restricted to POSIX local filesystems. No cross-platform
+  or network-filesystem guarantees were inferred from the Linux test run.
+
+No product service, marketplace, automatic model decisions, or DUD3-specific
+remediation catalog was added. No publication license was selected on the user's
+behalf; choose one before any public distribution.

vinyasar-0.3.0/MANIFEST.in

@@ -0,0 +1,8 @@
+include src/vinyasar/profile.schema.json
+include src/vinyasar/run.schema.json
+include CHANGELOG.md
+recursive-include tests *.py
+recursive-include docs *.md *.txt *.xml
+
+include src/vinyasar/run.v1.schema.json
+recursive-include tests/fixtures *.whl

vinyasar-0.3.0/PKG-INFO

@@ -0,0 +1,115 @@
+Metadata-Version: 2.4
+Name: vinyasar
+Version: 0.3.0
+Summary: Policy-driven, evidence-bound self-healing supervision for Yasarda (unified engine).
+Keywords: automation,self-healing,agents,yasarda,mcp,codemod,remediation,repository
+Classifier: Development Status :: 3 - Alpha
+Classifier: Environment :: Console
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Topic :: Software Development :: Version Control :: Git
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: jsonschema<5,>=4.23
+Provides-Extra: yaml
+Requires-Dist: PyYAML<7,>=6.0.2; extra == "yaml"
+Provides-Extra: mcp
+Provides-Extra: dev
+Requires-Dist: pytest<10,>=8; extra == "dev"
+
+# Vinyasar 0.2.1
+
+A bounded policy-driven supervisor for Yasarda, not an autonomous code writer.
+It has one mutating playbook: recover an interrupted Yasarda transaction to its
+original file state. Recovery does not finish applying the interrupted patch.
+
+The same Controller serves Python SDK, CLI and MCP. Models can propose changes,
+but cannot create playbooks, change pinned roots or grant authority through MCP.
+
+## Install
+
+Install the supplied Yasarda 0.3.1 and Vinyasar 0.2.1 wheels into a dedicated
+virtual environment. This delivery is local; it has not been published to PyPI.
+
+```sh
+python -m pip install ./yasarda-0.3.1-py3-none-any.whl ./vinyasar-0.2.0-py3-none-any.whl
+vinyasar --version
+vinyasar validate /private/profile.json
+vinyasar inspect /private/profile.json --trigger-id review-001
+```
+
+CLI `inspect` produces a dry-run record in the configured state/report storage;
+it does not repair the target. MCP `inspect_target` only inspects, without writing
+reports. Review the profile's authority before invoking `vinyasar run`: a generated
+profile can enable automatic recovery when setup checks consider storage eligible.
+
+```sh
+vinyasar run /private/profile.json --trigger manual --trigger-id event-001
+vinyasar-mcp --config /private/profile.json
+```
+
+Reusing the same trigger replays the same run. Use a new trigger for a new health
+check. All cooperating processes must share the same canonical state roots.
+
+## Changed in this version
+
+Fresh verification requires complete detector coverage and an exact rolled-back
+receipt. Recovery preflights file counts and byte limits in Yasarda itself.
+Durable run intent precedes attempt reservation; reserved uncertain actions are
+not blindly reissued after a crash. Whole-target locks serialize local workers.
+
+Retry budgets survive cosmetic profile edits. Report storage is checked before
+actions. Canonical reports are staged, fsynced and published without replacement,
+with recovery for recognized staging/link crash windows. Report schema v2 adds
+record integrity, final issues, inspection completeness and deadline status.
+Diagnostics are bounded/redacted before durable serialization. Failed MCP
+reconciliation returns isError=true; replay serialization is deterministic.
+
+## Upgrades and limits
+
+Read [docs/UPGRADE.md](docs/UPGRADE.md) before replacing a running installation.
+V1 reports are historical only. Legacy ledgers require host-only explicit
+adoption. Original Yasarda 0.2.0 journals require matching-engine recovery first.
+
+Linux/Python 3.13.5 was exercised for this iteration. Local POSIX mutation only;
+no Windows mutation, macOS validation, distributed lock or same-user sandbox is
+claimed. Timing is a cooperative deadline, not preemption of blocking filesystem
+calls. Redaction is bounded allowlisting, not universal secret detection.
+
+No scheduler installation, AST repair, network exporter, shell playbook, MCP
+Tasks or A2A adapter was added. Builds and application performance require their
+own independently authorized workers and evidence. See SAFETY.md for the boundary.
+
+
+## Host initialization and pinned paths (0.2.1)
+
+`vinyasar init` is the explicit host setup boundary. It resolves supplied path
+aliases once and writes canonical absolute repository, Yasarda-state,
+Vinyasar-state and report paths into a **new** profile. This includes home and
+YASARDA_STATE_DIR defaults. Missing state directories are not created; the nearest
+existing prefix is resolved strictly so dangling links and loops are refused.
+The profile output parent must already exist and be private. Its final component
+is never followed if linked and existing files are never overwritten.
+
+Loading an existing profile does not canonicalize links away. Both the profile
+file path and its configured paths must be link-free; relative configured entries
+remain compatible and are interpreted relative to the canonical profile parent.
+A linked `/var/...` source profile must be supplied explicitly as its canonical
+`/private/var/...` path. YAML uses the same bounded regular-file loader.
+
+`vinyasar doctor paths PROFILE` (or `vinyasar.doctor_paths(PROFILE)`) reports supplied
+and canonical path names, missing locations, and private-directory permission
+issues as JSON. It writes nothing, does not instantiate Controller, and never
+changes modes or follows a linked profile to read it. Exit 0 means these path and
+configuration checks passed; exit 2 reports a problem. The diagnostic explicitly
+sets `mutation_authorized: false` and `storage_eligibility: not_assessed`.
+It is not a sandbox, lock, filesystem certification, or approval to repair.
+
+Python path semantics reference: https://docs.python.org/3.13/library/pathlib.html
+(`Path.resolve(strict=True)`, strict loop and missing-path handling). Descriptor
+read guards reference: https://docs.python.org/3/library/os.html (`dir_fd`,
+`O_NOFOLLOW`, `O_NONBLOCK`, `fstat`). Runtime mutation/storage guards are unchanged.

vinyasar-0.3.0/README.md

@@ -0,0 +1,92 @@
+# Vinyasar 0.2.1
+
+A bounded policy-driven supervisor for Yasarda, not an autonomous code writer.
+It has one mutating playbook: recover an interrupted Yasarda transaction to its
+original file state. Recovery does not finish applying the interrupted patch.
+
+The same Controller serves Python SDK, CLI and MCP. Models can propose changes,
+but cannot create playbooks, change pinned roots or grant authority through MCP.
+
+## Install
+
+Install the supplied Yasarda 0.3.1 and Vinyasar 0.2.1 wheels into a dedicated
+virtual environment. This delivery is local; it has not been published to PyPI.
+
+```sh
+python -m pip install ./yasarda-0.3.1-py3-none-any.whl ./vinyasar-0.2.0-py3-none-any.whl
+vinyasar --version
+vinyasar validate /private/profile.json
+vinyasar inspect /private/profile.json --trigger-id review-001
+```
+
+CLI `inspect` produces a dry-run record in the configured state/report storage;
+it does not repair the target. MCP `inspect_target` only inspects, without writing
+reports. Review the profile's authority before invoking `vinyasar run`: a generated
+profile can enable automatic recovery when setup checks consider storage eligible.
+
+```sh
+vinyasar run /private/profile.json --trigger manual --trigger-id event-001
+vinyasar-mcp --config /private/profile.json
+```
+
+Reusing the same trigger replays the same run. Use a new trigger for a new health
+check. All cooperating processes must share the same canonical state roots.
+
+## Changed in this version
+
+Fresh verification requires complete detector coverage and an exact rolled-back
+receipt. Recovery preflights file counts and byte limits in Yasarda itself.
+Durable run intent precedes attempt reservation; reserved uncertain actions are
+not blindly reissued after a crash. Whole-target locks serialize local workers.
+
+Retry budgets survive cosmetic profile edits. Report storage is checked before
+actions. Canonical reports are staged, fsynced and published without replacement,
+with recovery for recognized staging/link crash windows. Report schema v2 adds
+record integrity, final issues, inspection completeness and deadline status.
+Diagnostics are bounded/redacted before durable serialization. Failed MCP
+reconciliation returns isError=true; replay serialization is deterministic.
+
+## Upgrades and limits
+
+Read [docs/UPGRADE.md](docs/UPGRADE.md) before replacing a running installation.
+V1 reports are historical only. Legacy ledgers require host-only explicit
+adoption. Original Yasarda 0.2.0 journals require matching-engine recovery first.
+
+Linux/Python 3.13.5 was exercised for this iteration. Local POSIX mutation only;
+no Windows mutation, macOS validation, distributed lock or same-user sandbox is
+claimed. Timing is a cooperative deadline, not preemption of blocking filesystem
+calls. Redaction is bounded allowlisting, not universal secret detection.
+
+No scheduler installation, AST repair, network exporter, shell playbook, MCP
+Tasks or A2A adapter was added. Builds and application performance require their
+own independently authorized workers and evidence. See SAFETY.md for the boundary.
+
+
+## Host initialization and pinned paths (0.2.1)
+
+`vinyasar init` is the explicit host setup boundary. It resolves supplied path
+aliases once and writes canonical absolute repository, Yasarda-state,
+Vinyasar-state and report paths into a **new** profile. This includes home and
+YASARDA_STATE_DIR defaults. Missing state directories are not created; the nearest
+existing prefix is resolved strictly so dangling links and loops are refused.
+The profile output parent must already exist and be private. Its final component
+is never followed if linked and existing files are never overwritten.
+
+Loading an existing profile does not canonicalize links away. Both the profile
+file path and its configured paths must be link-free; relative configured entries
+remain compatible and are interpreted relative to the canonical profile parent.
+A linked `/var/...` source profile must be supplied explicitly as its canonical
+`/private/var/...` path. YAML uses the same bounded regular-file loader.
+
+`vinyasar doctor paths PROFILE` (or `vinyasar.doctor_paths(PROFILE)`) reports supplied
+and canonical path names, missing locations, and private-directory permission
+issues as JSON. It writes nothing, does not instantiate Controller, and never
+changes modes or follows a linked profile to read it. Exit 0 means these path and
+configuration checks passed; exit 2 reports a problem. The diagnostic explicitly
+sets `mutation_authorized: false` and `storage_eligibility: not_assessed`.
+It is not a sandbox, lock, filesystem certification, or approval to repair.
+
+Python path semantics reference: https://docs.python.org/3.13/library/pathlib.html
+(`Path.resolve(strict=True)`, strict loop and missing-path handling). Descriptor
+read guards reference: https://docs.python.org/3/library/os.html (`dir_fd`,
+`O_NOFOLLOW`, `O_NONBLOCK`, `fstat`). Runtime mutation/storage guards are unchanged.

vinyasar-0.3.0/docs/ARCHITECTURE.md

@@ -0,0 +1,28 @@
+# Vinyasar 0.2.0 architecture
+
+Validated pinned profile -> target run lock -> canonical replay or durable run
+checkpoint -> complete deterministic inspection -> bounded policy decision ->
+durable action intent -> persistent attempt reservation -> reviewed playbook ->
+public Yasarda action -> matching receipt and fresh inspection -> terminal run
+checkpoint -> canonical JSON -> deterministic Markdown/CloudEvents projection.
+
+`Controller` is the only reconciliation implementation. `YasardaTarget` implements
+the typed Target interface. Setup probes and engine calls stay in that adapter.
+`StateStore` owns stable incident budgets and explicit legacy adoption. `RunStore`
+owns request-bound intent and terminal checkpoints. `storage` owns descriptor-
+relative private records, bounded locks and no-overwrite staged publication.
+`diagnostics` bounds and redacts before durable records. `ReportWriter` validates
+schema-v2 identity/digests/references and projects Markdown. CLI/MCP delegate.
+
+Yasarda 0.3.1 adds resource enforcement to its existing public recover operation;
+it retains plan compatibility marker 0.2.1. The supervisor does not become a
+second file transformer. Profiles retain the v1alpha1 API with two optional byte
+limits. Reports explicitly version to schema 2 rather than reinterpreting v1.
+
+Terminal reports describe history. New trigger IDs request new inspections;
+identical IDs reconstruct or replay existing work. After reservation, replay can
+verify a completed receipt but cannot assume an unreturned action never ran.
+
+Only local cooperating workers are coordinated. Protocol adapters and exporters
+are views over records, never evidence authorizing another repair. Scheduler
+planning/exporters/extra playbooks remain future work, not present features.

vinyasar-0.3.0/docs/HANDOFF.md

@@ -0,0 +1,21 @@
+# Vinyasar implementation handoff
+
+Version 0.2.1 pairs with Yasarda >=0.3.1,<0.4. Read SAFETY.md and UPGRADE.md first.
+Original test modules are preserved unchanged; new test_review_regressions.py
+contains review reproductions and process/storage failure tests. Its original
+independent review module is retained in the release evidence; four fixture or
+contract adaptations are documented at the bundle root.
+
+Keep single Controller/public adapter boundaries, strict profiles, finite attempt
+budgets, durable intent and conservative unknown-state refusal. Do not silently
+promote v1 reports or alter old plan version markers. Models and protocol callers
+cannot supply arbitrary executable playbooks or expand authority.
+
+The user supplied Vector's macOS arm64/CPython 3.12.11 validation for the
+exact 0.2.0 baseline: 207 active tests and both MCP smoke tools passed, with a
+canonical-temp-path workaround. This is reported external evidence, not a local
+rerun. The modified 0.2.1 release still needs its own native macOS run.
+Next bounded work: tools/validate_platform.py on macOS and preserved-journal
+upgrade rehearsal. Do not carry baseline platform qualification to changed code. Do not install schedulers,
+publish packages, push repositories, choose licenses or run cloud mutations based
+on this document. The bundle AGENT_HANDOFF.json describes current evidence.

vinyasar-0.3.0/docs/SAFETY.md

@@ -0,0 +1,108 @@
+# Vinyasar 0.2.1 safety contract
+
+## Authority and verification
+
+One Controller owns reconciliation. A validated host profile, built-in playbook
+and runtime capability are all required. The only mutating playbook recovers an
+interrupted Yasarda transaction. Engine calls go through yasarda_adapter; no
+production supervisor imports transaction internals. Test fixtures alone use
+private journal helpers to simulate interrupted publication.
+
+MCP retains its four tools; callers cannot choose roots, policy, playbooks or
+approvals. CLI/SDK are privileged host interfaces. The new adopt-ledger command
+requires explicit host confirmation and is not an MCP tool.
+
+Verification requires complete fresh pending-operation inspection, absence of
+the original incident, and a matching receipt with state rolled_back. A final
+inspection can revoke provisional success. INSPECTION_FAILED is never absence
+proof. Unresolved IDs have final issue/evidence records. None of this proves
+application health, build success or performance.
+
+## Durable execution
+
+Whole-target locks serialize cooperating workers at the same state location.
+Attempts use a stable target/recovery-store namespace, not cosmetic profile
+metadata. Incident/run locks have bounded waits. Digest-checked budget ledgers
+reject accidental counter corruption. Durable intent precedes reservation;
+reservation tokens prevent double accounting. A crash after reservation consumes
+an attempt even when the action has not yet started.
+
+Replay never blindly repeats an uncertain action. It checks the public Yasarda
+receipt and current target to reconstruct completed recovery; otherwise it records
+the uncertainty/failure. A new trigger may retry within the remaining budget.
+Completed records are checkpointed before final report publication. Replaying a
+completed report returns historical evidence, not a new health assessment.
+
+## Resources
+
+Yasarda preflights the full recovery snapshot: file count, per-file bytes and
+total bytes on each side. Exceeding a limit refuses the entire restoration.
+Observation counts, diagnostic fields, serialized inspections and reports are
+bounded before persistence. Canonical write/read share a 1 MiB record limit.
+
+Runtime is a cooperative deadline: bound lock waits, stop admitting work, report
+observed overruns. Blocking target adapters, filesystem calls and final report
+fsync can outlive it. No unsafe mid-transaction kill is used to fake a hard limit.
+
+## Storage and reporting
+
+State/report directories are owner-only, same-user, canonical POSIX directories.
+Ancestors are opened without following links. Existing permissions are not
+silently changed. Read guards reject links, nonregular files, unsafe permissions
+and oversized content. The only allowed two-link exception is an exact same-inode
+canonical/staging pair left by the documented publication crash window.
+
+Canonical bytes are fsynced in staging, linked to an absent destination, then
+staging is unlinked and the directory synchronized. Only exact bytes or an exact
+prefix in the reserved staging name may be resumed. Unknown bytes are preserved
+and refused. JSON precedes Markdown; missing identical Markdown can be regenerated
+on replay. Conflicting Markdown is not overwritten. No retention/cleanup job was
+added; recognized staging cleanup is part of its own publication protocol.
+
+Reports validate record/evidence digests, request binding and internal references.
+Checkpoints are separate from canonical reports. Required storage is preflighted
+before actions, but later disk failure remains possible. Checkpoints and underlying
+receipts retain the information needed for a subsequent same-trigger replay.
+
+## Privacy and explicit non-guarantees
+
+Free-form diagnostics and arbitrary facts are redacted/bounded before checkpoints
+or reports. Explicit identifiers remain visible; never put secrets in identifiers.
+This is not a universal secret scanner; hashes can reveal guessable low-entropy
+values. With redaction disabled, only bounded diagnostic content is retained.
+
+Digests detect corruption, not a writer able to rewrite content and digests. No
+hostile same-user containment, distributed coordination, power-loss proof,
+cryptographic authenticity or filesystem-wide compare-and-swap is claimed.
+Storage checks identify known unsupported locations but do not certify every
+accepted filesystem. Preserve state and keep external writers out.
+
+
+## Host initialization and pinned paths (0.2.1)
+
+`vinyasar init` is the explicit host setup boundary. It resolves supplied path
+aliases once and writes canonical absolute repository, Yasarda-state,
+Vinyasar-state and report paths into a **new** profile. This includes home and
+YASARDA_STATE_DIR defaults. Missing state directories are not created; the nearest
+existing prefix is resolved strictly so dangling links and loops are refused.
+The profile output parent must already exist and be private. Its final component
+is never followed if linked and existing files are never overwritten.
+
+Loading an existing profile does not canonicalize links away. Both the profile
+file path and its configured paths must be link-free; relative configured entries
+remain compatible and are interpreted relative to the canonical profile parent.
+A linked `/var/...` source profile must be supplied explicitly as its canonical
+`/private/var/...` path. YAML uses the same bounded regular-file loader.
+
+`vinyasar doctor paths PROFILE` (or `vinyasar.doctor_paths(PROFILE)`) reports supplied
+and canonical path names, missing locations, and private-directory permission
+issues as JSON. It writes nothing, does not instantiate Controller, and never
+changes modes or follows a linked profile to read it. Exit 0 means these path and
+configuration checks passed; exit 2 reports a problem. The diagnostic explicitly
+sets `mutation_authorized: false` and `storage_eligibility: not_assessed`.
+It is not a sandbox, lock, filesystem certification, or approval to repair.
+
+Python path semantics reference: https://docs.python.org/3.13/library/pathlib.html
+(`Path.resolve(strict=True)`, strict loop and missing-path handling). Descriptor
+read guards reference: https://docs.python.org/3/library/os.html (`dir_fd`,
+`O_NOFOLLOW`, `O_NONBLOCK`, `fstat`). Runtime mutation/storage guards are unchanged.

vinyasar-0.3.0/docs/TEST_REPORT.md

@@ -0,0 +1,6 @@
+# Validation for this source revision
+
+73 source tests pass in the resumed session on Linux/Python 3.13.5. Original
+release evidence is clearly separated under baseline/. See the complete bundle's
+docs/TEST_REPORT.md for final installed-wheel, subprocess, artifact and patch
+checks. This source result is not a claim about any other platform or runtime.

vinyasar-0.3.0/docs/UPGRADE.md

@@ -0,0 +1,72 @@
+# Upgrade without losing recovery history
+
+## Yasarda 0.3.0 to 0.3.1
+
+Stop cooperating writers first. Retain the same repository and Yasarda state
+paths. The plan compatibility marker remains 0.2.1; compatible journals retain
+identity. Host file/byte limits can now refuse previously unbounded recovery.
+Review the complete footprint rather than blindly widening limits.
+
+## Original Yasarda 0.2.0
+
+The originally delivered 0.2.0 wheel has a different plan identity marker. The
+new engine deliberately refuses those journals. Never edit `engine_version`,
+regenerate a journal digest, or relabel an old plan to make it pass validation.
+
+Use an isolated environment with the exact trusted old wheel. Review every
+pending operation with that version's doctor and receipt commands, then invoke
+its guarded recovery with explicit approval. For example:
+
+```sh
+/path/to/old-env/bin/yasarda --repo /canonical/local/repo \
+  --state-dir /private/old-yasarda-state doctor
+/path/to/old-env/bin/yasarda --repo /canonical/local/repo \
+  --state-dir /private/old-yasarda-state recover OPERATION_ID \
+  --approve OPERATION_ID --yes
+```
+
+Verify the returned receipt and repository bytes, including refusal on newer
+edits. Resolve every old pending operation before switching. Retain the old
+journals for the matching engine to read. Only then point the new installation
+at an explicitly chosen, separate new Yasarda journal directory.
+
+The regression fixture performs real old-engine recovery, retains its completed
+journal, and observes the restored repository through the new engine and new
+state directory. This is a tested operator procedure, not automatic migration.
+
+## Vinyasar 0.1.0 to 0.2.0
+
+Pause callers/schedulers and preserve the original profiles, reports and ledgers.
+Use a distinct report directory for schema-v2 records. Keep the supervisor state
+directory so retry history cannot silently disappear. Authoritative v2 replay
+refuses v1 reports; `run.v1.schema.json` is retained for historical inspection.
+
+Legacy retry ledgers lack reliable target identity. The host must identify the
+old profile revisions that belong to this target. Import each exact revision:
+
+```sh
+vinyasar adopt-ledger /private/profile.json \
+  --legacy-revision EXACT_64_CHARACTER_REVISION --yes
+```
+
+This preserves the old file, pins its digest, conservatively merges counters,
+and refuses a changed adopted file. Unrecognized legacy ledgers block actions.
+This command is not exposed by MCP. Do not guess ownership when unrelated targets
+shared a legacy state directory; resolve that configuration explicitly first.
+
+Use fresh event IDs for new scheduler events. Reusing a completed trigger requests
+historical replay, not a fresh health check. After an interrupted 0.2.0 run, retry
+its identical trigger to reconstruct and finish reporting its actual history.
+An indeterminate reserved action is not reissued by that replay; a separate new
+trigger may attempt recovery within the remaining budget.
+
+## Coordination and scope
+
+Cooperating workers must use the same canonical target and state roots. Changing
+those roots changes coordination identity. Do not select empty state merely to
+bypass an exhausted budget or pending operation. Keep unrelated writers out.
+
+State/report directories must be private and owned by the executing user. Wrong
+permissions and links are refused rather than silently chmodded or followed.
+Run mutation tests only in disposable local POSIX directories, never a cloud
+synced user checkout. No completed v1 report gains v2 guarantees retroactively.

vinyasar-0.3.0/docs/baseline/TEST_REPORT.md

@@ -0,0 +1,34 @@
+# Vinyasar 0.1.0 verification report
+
+Date: 2026-09-09
+Platform: macOS 26.6.2 arm64
+Python: 3.14.6
+
+## Results
+
+| Run | Result |
+| --- | --- |
+| Source suite | 20 passed, 0 failed, 0 skipped |
+| Ruff static checks | Passed |
+| Ruff format check | Passed |
+| Wheel and source distribution build | Passed |
+| Installed wheel, Python 3.14 clean environment | 20 passed, 0 failed, 0 skipped |
+| Installed wheel, Python 3.12 clean environment | 20 passed, 0 failed, 0 skipped |
+| Installed `vinyasar --version` | Passed; version 0.1.0 |
+| Installed `vinyasar-mcp --help` | Passed |
+| Installed generated-profile validate/inspect/run smoke | Passed |
+| Installed dependency check | All packages compatible |
+
+Coverage includes profile generation/validation, duplicate and overlapping
+configuration rejection, monotonic policy decisions, evidence-bound approvals,
+unknown-issue escalation, persistent cooldown and attempt budgets, dry runs,
+idempotent trigger replay, append-only reports, CloudEvents envelopes, fixed-scope
+MCP arguments, explicit journal-state pinning, three-store filesystem gating, and
+a real guarded recovery through Yasarda's public Engine API.
+The installed suites ran with an empty `PYTHONPATH`; imports resolved from each
+clean virtual environment's `site-packages` directory. A setup probe against the
+actual iCloud Desktop generated escalation-only recovery policy as designed.
+
+No test claims hardware power-loss safety, hostile same-user isolation, remote or
+cloud filesystem mutation safety, semantic application correctness, or
+cryptographic receipt authenticity.

vinyasar-0.3.0/docs/baseline/installed-environment.txt

@@ -0,0 +1,7 @@
+Platform: macOS 26.6.2 arm64
+Python 3.14: 3.14.6
+Python 3.12: 3.12.13
+Yasarda: 0.3.0
+Vinyasar: 0.1.0
+jsonschema: 4.26.0
+pytest: 9.1.1

vinyasar-0.3.0/docs/baseline/installed-wheel-test-results.txt

@@ -0,0 +1 @@
+20 passed in 0.26s

vinyasar-0.3.0/docs/baseline/installed-wheel-test-results.xml

@@ -0,0 +1 @@
+<?xml version="1.0" encoding="utf-8"?><testsuites name="pytest tests"><testsuite name="pytest" errors="0" failures="0" skipped="0" tests="20" time="0.166" timestamp="2026-09-09T12:26:56.609738-05:00" hostname="centaur-macmini-1"><testcase classname="tests.test_cli" name="test_cli_generates_valid_profile_without_overwrite" time="0.005" /><testcase classname="tests.test_config" name="test_generated_profile_is_normalized_and_revision_pinned" time="0.002" /><testcase classname="tests.test_config" name="test_profile_rejects_overlapping_rules" time="0.002" /><testcase classname="tests.test_config" name="test_profile_rejects_report_storage_inside_target" time="0.002" /><testcase classname="tests.test_config" name="test_load_rejects_duplicate_json_keys" time="0.001" /><testcase classname="tests.test_config" name="test_observe_rule_cannot_smuggle_capability" time="0.002" /><testcase classname="tests.test_config" name="test_profile_requires_a_pinned_yasarda_state_directory" time="0.003" /><testcase classname="tests.test_config" name="test_generated_profile_escalates_when_automation_state_is_cloud_synced" time="0.001" /><testcase classname="tests.test_controller" name="test_healthy_run_writes_append_only_reports" time="0.003" /><testcase classname="tests.test_controller" name="test_exact_approval_can_resume_on_a_new_trigger" time="0.004" /><testcase classname="tests.test_controller" name="test_dry_run_never_invokes_playbook" time="0.002" /><testcase classname="tests.test_controller" name="test_real_yasarda_pending_operation_is_recovered" time="0.099" /><testcase classname="tests.test_controller" name="test_events_are_cloudevents_compatible" time="0.005" /><testcase classname="tests.test_controller" name="test_volatile_evidence_does_not_reset_incident_attempt_budget" time="0.008" /><testcase classname="tests.test_mcp" name="test_mcp_surface_is_fixed_and_reconciles" time="0.003" /><testcase classname="tests.test_mcp" name="test_mcp_rejects_caller_scope_arguments" time="0.001" /><testcase classname="tests.test_policy" name="test_approval_id_is_exact_and_evidence_bound" time="0.002" /><testcase classname="tests.test_policy" name="test_mismatched_capability_escalates" time="0.002" /><testcase classname="tests.test_policy" name="test_unknown_issue_cannot_gain_action_authority" time="0.001" /><testcase classname="tests.test_state" name="test_attempt_budget_and_cooldown_are_persistent" time="0.002" /></testsuite></testsuites>

vinyasar-0.3.0/docs/baseline/source-test-results.txt

@@ -0,0 +1 @@
+20 passed in 0.17s

vinyasar-0.3.0/docs/baseline/source-test-results.xml

@@ -0,0 +1 @@
+<?xml version="1.0" encoding="utf-8"?><testsuites><testsuite name="pytest" errors="0" failures="0" skipped="0" tests="20" time="0.171" timestamp="2026-09-09T12:25:50.989915-05:00" hostname="centaur-macmini-1"><testcase classname="tests.test_cli" name="test_cli_generates_valid_profile_without_overwrite" time="0.006" /><testcase classname="tests.test_config" name="test_generated_profile_is_normalized_and_revision_pinned" time="0.001" /><testcase classname="tests.test_config" name="test_profile_rejects_overlapping_rules" time="0.002" /><testcase classname="tests.test_config" name="test_profile_rejects_report_storage_inside_target" time="0.002" /><testcase classname="tests.test_config" name="test_load_rejects_duplicate_json_keys" time="0.001" /><testcase classname="tests.test_config" name="test_observe_rule_cannot_smuggle_capability" time="0.002" /><testcase classname="tests.test_config" name="test_profile_requires_a_pinned_yasarda_state_directory" time="0.002" /><testcase classname="tests.test_config" name="test_generated_profile_escalates_when_automation_state_is_cloud_synced" time="0.001" /><testcase classname="tests.test_controller" name="test_healthy_run_writes_append_only_reports" time="0.004" /><testcase classname="tests.test_controller" name="test_exact_approval_can_resume_on_a_new_trigger" time="0.004" /><testcase classname="tests.test_controller" name="test_dry_run_never_invokes_playbook" time="0.002" /><testcase classname="tests.test_controller" name="test_real_yasarda_pending_operation_is_recovered" time="0.110" /><testcase classname="tests.test_controller" name="test_events_are_cloudevents_compatible" time="0.003" /><testcase classname="tests.test_controller" name="test_volatile_evidence_does_not_reset_incident_attempt_budget" time="0.006" /><testcase classname="tests.test_mcp" name="test_mcp_surface_is_fixed_and_reconciles" time="0.002" /><testcase classname="tests.test_mcp" name="test_mcp_rejects_caller_scope_arguments" time="0.001" /><testcase classname="tests.test_policy" name="test_approval_id_is_exact_and_evidence_bound" time="0.002" /><testcase classname="tests.test_policy" name="test_mismatched_capability_escalates" time="0.002" /><testcase classname="tests.test_policy" name="test_unknown_issue_cannot_gain_action_authority" time="0.001" /><testcase classname="tests.test_state" name="test_attempt_budget_and_cooldown_are_persistent" time="0.002" /></testsuite></testsuites>