ailoop v0.2.8

ailoop v0.2.8

Fixes

• Retry admission now serializes with durable stop mutations. A stop committed before admission
  suppresses the retry, while a later stop still cancels admitted work through the normal active-run
  path.
• Stop intent now remains dominant over later pause requests while explicit resume continues to run
  remaining work.

Security

• State lock handling now pins operations to owner-controlled, non-group/world-writable state
  namespaces; rejects unsafe path types, ownership, hardlinks, and changed lock identities; normalizes
  .locks and lock files to modes 0700 and 0600; and writes diagnostic PIDs only after successful
  lock validation.
• The configured state-root ancestry, processes sharing the effective user ID, and non-local
  filesystem lock semantics remain trusted boundaries rather than claimed isolation guarantees.

Compatibility

• prompt.pre_prompt_enabled, prompt.attach_agent_file, loop.continue_on_error, and
  tasks.stop_when_complete now require values that PyYAML parses as booleans. Quoted strings,
  numbers, null, and collections are rejected with field-specific configuration errors.
• Use unquoted YAML true and false for portable, explicit configuration. PyYAML legacy spellings
  that already parse as booleans remain accepted.

Quality

• The release-preparation suite passes 506 tests with 85% coverage, including adversarial lock-path,
  retry-control, explicit-resume, and strict-config regressions.

Install

pipx install ./ai_loop-0.2.8-py3-none-any.whl
ailoop --help

PyPI publication is not part of this release.

ailoop v0.2.7

ailoop v0.2.7

Fixes

• Programmatic loop-table rebuilds now suppress DataTable.RowHighlighted, so refreshes no longer
  rebind transient rows or clear unsaved loop drafts, selection, cursor position, or workspace
  branch state.

Quality

• Added Python 3.11 CI regression coverage for the refresh and draft-preservation path.

Install

pipx install ./ai_loop-0.2.7-py3-none-any.whl
ailoop --help

PyPI publication is not part of this release.

ailoop v0.2.6

ailoop v0.2.6

Fixes

• Lock teardown and file-close failures no longer turn an already committed state mutation into a
  reported operation failure.
• Best-effort postcommit event suppression is limited to storage OSError failures; programming
  errors and interrupts still surface.
• Ambiguous append-succeeded-then-raised outcomes do not retry or duplicate events, workers, or
  consumed follow-up intent.

Quality

• 418 tests pass with 85% coverage on the corrective release line.
• Python 3.11 and Python 3.13 CI, Ruff, package metadata checks, clean-wheel installation, and
  downloaded checksum verification are required before completion.

Install

pipx install ./ai_loop-0.2.6-py3-none-any.whl
ailoop --help

PyPI publication is not part of this release.

ailoop v0.2.5

ailoop v0.2.5

Reliability

• Memory IDs are collision-safe and publication cannot overwrite an existing entry.
• Arbitrary runner output bytes remain exact on disk and render safely in summaries, CLI, and TUI.
• Committed run intents remain successful when best-effort event journaling fails.
• Retry and reset executions preserve distinct prompt, stdout, and stderr evidence.
• Concurrent workspace-history appends and compaction are serialized and atomic.

TUI safety

• Hidden or editor-focused memory shortcuts fail closed.
• Loop actions require a visible, stable row identity; hidden query drafts cannot mutate loops.
• Memory archive and delete confirmations are bound to the same entry ID across both keypresses.
• Detached worker and replay feedback reports launch requests and recoverable launch failures honestly.

Quality

• Release validation includes 405 passing tests with 85% coverage, Ruff, package checks,
  clean-wheel installation, Python 3.11 and Python 3.13 CI, and downloaded artifact checksum
  verification.

Install

Download the wheel attached to this GitHub release, then run:

pipx install ./ai_loop-0.2.5-py3-none-any.whl
ailoop --help

PyPI publication is not part of this release.

ailoop v0.2.4

ailoop v0.2.4

Fixes

• Configured loop creation now writes complete state atomically and removes a newly created loop if
  creation-event or workspace-history persistence fails. (#104)
• Scheduled loops now fail closed while a worker is active: follow-up metadata changes are rejected,
  emergency stop is the only claimed-process control, and cleanup failures are refresh-only. (#104)
• Scheduled draft, branch strategy, safety, and notification surfaces now describe saved planning
  metadata honestly; unsupported controls are disabled and excluded from keyboard focus. (#104)

Safety

• Recoverable service and worker-spawn failures now report the error, refresh state, and avoid false
  success messages. Unsaved follow-up text is preserved when a worker cannot start. (#104)
• Start, resume, follow-up, and next-iteration side effects occur only after their corresponding
  persistence or spawn step succeeds. (#104)

Quality

• Added regressions for creation rollback, fail-closed scheduled edits, compact-layout capability
  copy, planning-only focus behavior, recoverable mutations and spawns, and input preservation. (#104)
• Release validation: 319 tests passed with 85% coverage; Ruff, diff, compile, and lock checks
  passed; Python 3.11 and Python 3.13 PR CI passed; wheel and source distributions passed
  twine check; and the wheel installed and reported version 0.2.4 in a clean environment.

Install

Download the wheel attached to this GitHub release, then run:

pipx install ./ai_loop-0.2.4-py3-none-any.whl
ailoop --help

PyPI publication is not part of this release.

ailoop v0.2.3

ailoop v0.2.3

Fixes

• Post-run task-file validation failures now emit a dedicated truthful event while preserving the
  original validation exception as the primary error. Secondary history/event failures are attached
  as notes instead of masking it. (#101)
• Saved scheduled configurations can transition atomically to fixed or infinite execution, while
  requests that remain scheduled are still rejected from manual restart. (#102)
• TUI restart and worker-spawn failures are reported without escaping the action handler. Scheduled
  help now shows emergency stop only for a claimed process and never advertises deletion while the
  loop is active. (#102)

Safety

• Deferred validation catches only expected OSError/ValueError failures; interrupts and process
  exits retain their normal control-flow semantics.
• Completed iteration accounting, output artifacts, and resume numbering remain durable across
  task-file validation failures.
• A failed worker spawn leaves the converted executable configuration persisted and idle, with an
  explicit recoverable error message.

Quality

• Added validation-primary combined-failure, deleted-task-file, resume-artifact, scheduled-mode
  transition, spawn-failure, emergency-stop-help, and delete-help regressions.
• Release validation: 293 tests passed with 83% coverage; Ruff and lock checks passed; wheel
  and source distributions passed twine check; and the wheel installed and reported version
  0.2.3 in a clean virtual environment.

Install

Download the wheel attached to this GitHub release, then run:

pipx install ./ai_loop-0.2.3-py3-none-any.whl
ailoop --help

PyPI publication is not part of this release.

ailoop v0.2.2

ailoop v0.2.2

Fixes

• Completed iteration accounting now survives post-run task-file validation failures, so a
  successful runner is not repeated after the task file is repaired. (#97)
• memory edit patches only explicitly supplied fields against the latest snapshot instead of
  resetting omitted values to current application defaults. (#98)
• Scheduled loops are presented honestly as saved configurations while scheduler execution is
  unavailable. Unsupported runtime controls are disabled and guarded. (#99)

Safety

• Finalized iteration events and workspace history are persisted before post-run validation errors
  surface.
• Memory updates use stable per-entry locks and durable atomic replacement; concurrent edits and
  usage updates no longer lose each other.
• Scheduled follow-ups are saved without spawning a process, pending manual work cannot be
  overwritten by a schedule save, and emergency stop remains available for a claimed process.

Quality

• Added regressions for task-file failure recovery, memory patch/version integrity, concurrent
  memory mutation, atomic-write failures, scheduled keyboard actions, and scheduler-unavailable
  status copy.
• Release validation: 287 tests passed with 83% coverage; Ruff and lock checks passed; wheel
  and source distributions passed twine check; and the wheel installed and ran in a clean
  virtual environment.

Install

Download the wheel attached to this GitHub release, then run:

pipx install ./ai_loop-0.2.2-py3-none-any.whl
ailoop --help

PyPI publication is not part of this release.

ailoop v0.2.1

ailoop v0.2.1

Fixes

• ailoop resume now establishes an explicit run intent for paused and stopped loops.
• Loop state and control transitions are transactional, so newer pause or stop requests win
  before a detached dashboard worker can claim the next iteration.
• State writes use durable atomic replacement, while event appends are flushed and synced.

Safety

• On POSIX, catchable runner exits clean and reap the original process group before the loop
  becomes resumable.
• Unconfirmed cleanup records cleanup_failed and blocks rerun, resume, control, and removal;
  it does not claim that the process was terminated.
• Abrupt parent SIGKILL can leave a runner alive and require manual cleanup. The persisted loop
  remains active and fails closed instead of authorizing duplicate work.
• Removed loop IDs are durably retired and cannot be reused by delayed workers or events.
• Scheduled loops remain configuration-only until scheduler execution is implemented.

Quality

• Added lifecycle-safety coverage for control races, durable claims, cleanup failures,
  process-group teardown, retired IDs, removal races, and scheduled-loop guardrails.
• CI passes on Python 3.11 and 3.13.
• 272 tests pass with 83% coverage.

Install

Download the wheel attached to this GitHub release, then run:

pipx install ./ai_loop-0.2.1-py3-none-any.whl
ailoop --help

PyPI publication is not part of this release.

ailoop v0.2.0

ailoop v0.2.0

Adds

• Searchable TUI loop navigation across loop IDs, state, mode, runner, agent, prompt, workspace, and task-file metadata.
• A recent-workspace picker plus workspace-aware follow-up and history tracking.
• Runner iteration timeouts and more responsive stop behavior.

Changes

• TUI controls preserve unsaved editor input while searching and switching loops.
• Workspace-root validation and status are surfaced before spawning a runner.
• Loop refreshes reuse one state snapshot instead of repeatedly scanning state files.
• The Textual dependency now targets the tested 8.2 release line instead of an incompatible legacy floor.

Fixes

• Stop TUI refresh callbacks safely during shutdown.
• Observe pause and stop controls during loop delays.
• Interrupt active runners when a stop is requested.
• Validate workspace roots before spawning.
• Clarify queued follow-up and reset/restart behavior.

Quality

• Expanded TUI, service, prompting, and workspace-loop regression coverage.
• Added CI for Python 3.11 and 3.13.
• Enforced an 80% branch-coverage floor.

Install

Download the wheel attached to this GitHub release, then run:

pipx install ./ai_loop-0.2.0-py3-none-any.whl
ailoop --help

PyPI publication is not part of this release.

ailoop v0.1.0

ailoop v0.1.0

• Launches ailoop, a small CLI for running AI terminal tools in safe, repeatable loops.
• Adds YAML config, CLI overrides, and durable local state for resumable runs.
• Supports OpenCode, Codex, Claude, and other command-template runners.
• Includes strict Markdown task-file mode for structured iterative work.
• Ships pause, resume, stop, logs, stats, tail, and JSON output for automation.