trigger-tree v1.3.2

1.3.2 — 2026-07-20

• Restores Claude Code startup and /tt discovery after the withdrawn v1.3.1 package: shared hooks now use the Claude-compatible plugin-root variable that Codex also provides, and the Claude command has an explicit conventional skills/tt/SKILL.md.
• Prevents duplicate shared Claude events while retaining Claude-only skill, failure, and session-end telemetry. Codex-only Stop outcomes no longer run in Claude.
• Adds installed-package regression checks for the command skill and hook environment.

trigger-tree v1.3.0

1.3.0 — 2026-07-20

• Adds a native Codex plugin manifest, marketplace catalog, trigger-tree skill, and
  lifecycle hooks while retaining the existing Claude Code plugin in the same repo.
• Normalizes Codex unified-exec cmd payloads, native reads, and filesystem MCP reads
  into the shared telemetry schema. Repository-root resolution keeps one dataset when
  Codex starts in a nested directory; hook failures remain silent and non-blocking.
• Documents and tests both installation paths. The website remains exactly six USP
  cards in a three-column desktop grid.

trigger-tree v1.2.2

1.2.2 — 2026-07-20

• Keeps the v1.2.1 dedicated sort-legend fix while replacing the replay integration
  test's wall-clock race with a deterministic clock. Slow or suspended macOS runners
  can no longer miss the replay branch and report platform-specific 99.8% coverage.
• v1.2.1 was withdrawn because its tag CI was not fully green. v1.2.2 is the first

trigger-tree v1.2.0

A more legible, navigable live dashboard

• Historical prompts now show either a width-aware truncated preview, a stable hash ID, or an explicit text-off label—without weakening the privacy default.
• Heat indicators are horizontal five-cell bars, aligned against the pane's available right edge so wide splits expose more filename context.
• New explicit live sorts: f focus/recent, h hottest, c coldest including untouched, and n A–Z. The active mode stays visible in the footer.
• Scan-only activity remains useful in focus mode but never counts as heat or contaminates hot/cold ordering.
• README, plugin help, changelog, and the interactive website demo match the behavior; the website remains six cards in a fixed three-column desktop grid.

Verified by 115 tests at 100% coverage, Python 3.10–3.13, Ubuntu/macOS/Windows, formatting and shell gates, workflow-security audits, Claude plugin validation, isolated install, Pages, and release integrity.

trigger-tree v1.1.0

Temporal heat, durable evidence

• Heat now measures current attention with a 30-day half-life; lifetime reads never decay.
• Insights adds exact 7/30/90-day windows, last-read dates, and separate file/folder heat.
• The live watcher shows heat beside lifetime counts and reheats immediately on new reads.
• Full-screen rendering now disables autowrap and clears dashboard scrollback, then restores the terminal on exit.
• README, plugin instructions, and the six-card three-column website demo document the model and its safety boundary: cold means inactive now, never safe to remove.

Verified by 112 tests at 100% coverage, Python 3.10–3.13, Ubuntu/macOS/Windows, Black, Ruff, ShellCheck, workflow-security audits, Claude plugin validation, isolated installation, and tag/release integrity.

trigger-tree v1.0.4

Urgent zsh compatibility hotfix for runtime Bash-read capture.

• Replaces the wrapper variable status, which is read-only in zsh, with a private shell-neutral exit-code variable.
• Prevents successful documentation reader commands from being changed to exit 1.
• Restores expanded-path telemetry logging in zsh.
• Proves the exact PICK=$(pipeline); cat "$PICK" scenario in both bash and zsh, including unchanged stdout, exit code 0, loop-variable reads, and recorded paths.

Verified before publication: 111 tests, 100% Python coverage, Linux/macOS/Windows, Python 3.10/3.11/3.13, Black, Ruff, ShellCheck, workflow security, Claude plugin validation, isolated installation, and tag-to-metadata release integrity.

trigger-tree v1.0.3

Reliable Bash documentation reads after shell expansion.

• Captures successful cat/head/tail/non-mutating sed/awk reads after variables, command substitutions, loops, and globs resolve.
• Uses Claude Code session environment preambles; does not rewrite commands, alter permissions, or inspect output.
• Stores only normalized matching doc paths, never commands, patterns, output, or contents.
• Preserves stdout, stderr, arguments, and exit status.
• Keeps the conservative literal-path fallback where runtime Bash capture is unavailable.
• Prevents duplicate PostToolUse reads while runtime capture is active.
• Updates README, privacy documentation, website copy, and CI ShellCheck coverage.
• Keeps exactly six USP cards in a three-column desktop grid.

Verified before publication: exact PICK=$(...) + cat "$PICK" regression, loop-variable reads, 110 tests, 100% Python coverage, Linux/macOS/Windows, Python 3.10/3.11/3.13, Black, Ruff, ShellCheck, workflow security, Claude plugin validation, isolated installation, and tag-to-metadata release integrity.

trigger-tree v1.0.2

Telemetry visibility fixes for shell-driven documentation discovery.

• Statusline freshness now follows the newest read or scan event.
• Scan-only sessions remain visibly active, with a scan counter and folder-style trailing path.
• File/folder/depth totals remain derived from actual reads.
• Bash cat, head, tail, non-mutating sed, and awk calls now log existing watched docs as ordinary reads.
• Multiple doc arguments are captured; sed -i variants and non-doc files remain excluded.
• Cross-process coverage proves Bash scans and reads flow through the live watcher.

Verified before publication: 106 tests, 100% Python coverage, Linux/macOS/Windows, Python 3.10/3.11/3.13, Black, Ruff, ShellCheck, workflow security, Claude plugin validation, isolated marketplace installation, and tag-to-metadata release integrity.

trigger-tree v1.0.1

Safety-contract regression coverage.

• Adds an isolated critical-tag protection test with no glob or safety-path fallback.
• Adds an isolated widely-linked protection test at the three-in-link threshold.
• Both assert protected classification and the rare-but-critical likely-keep recommendation.

Verified before publication: 103 tests, 100% Python coverage, Linux/macOS/Windows, Python 3.10/3.11/3.13, formatting, linting, workflow security, Claude plugin validation, isolated marketplace installation, and tag-to-metadata release integrity.

trigger-tree v1.0.0

trigger-tree 1.0 is the stable foundation for trustworthy, local documentation-discovery telemetry.

Highlights:

• Recursive CLAUDE.md @import discovery for accurate always-loaded documentation signals
• Subagent read attribution and compaction-aware deduplication
• Safe review candidates with protection for rare, critical documentation
• Versioned telemetry schema with migration and future-version rejection
• Stronger /tt doctor checks for configuration, hooks, Python, and history
• Opt-in experimental outcome correlation

Verified before publication:

• 101 tests with 100% Python coverage
• Linux, macOS, and Windows
• Python 3.10, 3.11, and 3.13
• Black, Ruff, ShellCheck, actionlint, zizmor
• Claude plugin validation and isolated marketplace installation
• Tag-to-manifest release integrity