v0.1.5

Guard local manifests before release

Add a safer project-manifest boundary by treating private .tali manifests as local by default, while preserving explicit shared manifests and adding cached update visibility.

Constraint: Tali must not silently rewrite repository policy or make network checks part of run correctness.

Rejected: Always ignore .tali globally | would remove the shareable project-manifest path.

Rejected: Check for updates on every command without cache | would slow normal CLI usage and create noisy offline failures.

Confidence: high

Scope-risk: moderate

Directive: Keep post-run notices advisory and never let update/gitignore checks change manifest execution status.

Tested: cargo fmt --check; cargo test --locked --all-targets; cargo clippy --locked --all-targets -- -D warnings; cargo check --locked; release metadata, crates version, license, diff checks; manual private/shared manifest and update-check smokes

Not-tested: Full release workflow until tag push

v0.1.4

Steer agents toward Tali handoffs

The local Codex verification showed implicit one-command setup handoffs could still become ad-hoc shell scripts. Tighten the bundled skill trigger text and ship it as a patch release so installer-provided skills bias agents toward Tali manifests.

Constraint: Implicit skill selection depends heavily on compact skill descriptions when many skills are installed.

Rejected: Only documenting the behavior | documentation did not change the observed local Codex behavior.

Confidence: high

Scope-risk: narrow

Directive: Keep the Tali skill description short, trigger-rich, and explicit about not creating setup scripts as the primary handoff.

Tested: cargo fmt --check; cargo check --locked; cargo clippy --locked --all-targets -- -D warnings; cargo test --locked --all-targets; python3 scripts/check-release-metadata.py; python3 scripts/check-crates-version.py; python3 scripts/check-licenses.py; git diff --check; local Codex implicit handoff test; tali run setup --yes --input-env api_token=API_TOKEN_FOR_TALI_TEST

Not-tested: Windows interactive Codex client behavior

v0.1.3

Stabilize live log following before release

Handle transient run.json parse failures while a run is still writing logs so live observers do not fail during startup or shutdown races.

Constraint: Release validation exposed this through CI even though the v0.1.2 release workflow completed.

Rejected: test-only retry workaround | the same race can affect real tali logs follow users.

Confidence: high

Scope-risk: narrow

Directive: Treat partially written live-log state as in-progress, not as a hard failure.

Tested: cargo fmt --check; cargo check --locked; cargo clippy --locked --all-targets -- -D warnings; cargo test --locked --all-targets; release metadata/license/crate-version checks.

Not-tested: full release-check after this commit before clean-tree commit creation.

v0.1.2

Prepare Tali for safer handoffs

Add conservative cleanup and placeholder replacement so agent-authored manifests can manage long-lived logs and fill existing config templates without exposing rendered secrets.

Constraint: Tali must remain a predictable runtime, not a planner.

Rejected: automatic manifest or secret cleanup | manifests and secrets must not be deleted by routine cleanup.

Rejected: sequential replacement of inserted text | replacement should be based on the original file only.

Confidence: high

Scope-risk: moderate

Directive: Keep cleanup preview-first and keep rendered secret-bearing file contents out of logs.

Tested: cargo fmt --check; cargo check --locked; cargo clippy --locked --all-targets -- -D warnings; cargo test --locked --all-targets; release metadata/license/crate-version checks.

Not-tested: full release-check before commit because it requires a clean working tree.

v0.1.1

Make Tali easier for agents to hand off

Agents can now register manifests without exposing the add step to users, installs can refresh the bundled skill into explicit directories, and release installers can update an installed binary in place on Unix.

Constraint: User wanted a curl-installed public release with agent skill setup and one-command user handoff.

Rejected: Requiring users to run tali add manually | Agents should perform registration and show only tali <id>.

Confidence: high

Scope-risk: moderate

Directive: Keep Tali as a runtime; do not add planner behavior to satisfy handoff ergonomics.

Tested: cargo fmt --check; cargo test --locked --all-targets; cargo clippy --locked --all-targets -- -D warnings; quick_validate.py .codex/skills/tali-agent; sh -n scripts/install.sh; PowerShell parser check; actionlint; local file:// installer plus tali update smoke.

Not-tested: Windows self-update over the currently running executable.

v0.1.0

Release Tali 0.1.0