v4.10.18 — fix hook paths when launched from subdirectory

Hook commands now cd to the git root before running, so pre-bash.sh / post-bash.sh / post-write.sh work correctly when Claude Code is opened from a subdirectory (e.g. app/) rather than the project root. The hook scripts use relative paths internally (scripts/sdlc/, docs/stories/) and were silently broken in that case.

v4.10.17 — fix shellcheck in test_skills_sh.sh

Fix SC2015 (&&/|| anti-pattern) and SC2034 (unused variable) in tests/template/test_skills_sh.sh — CI was failing the lint step.

v4.10.16 — drop duplicate skills/ dir

Remove skills/ directory added in v4.10.15 — canonical skill location is template/.claude/skills/ (per harness). Root SKILL.md kept as the package descriptor for skills.sh.

v4.10.15 — skills.sh integration: tests, registry fallback, publishing

What's new

skills.sh integration overhaul

• Fixed npx skills invocation across the board: replaced --yes flag with npm_config_yes=true env var — the old approach failed in projects without a package.json, which is most non-Node projects
• [F] skills browser now reliably works from any project type

pipeline-runner registry fallback

• When a skill isn't found locally, pipeline-runner now tries npx skills add kinncj/maple@<name> from skills.sh before giving up
• Applies to both Claude Code and OpenCode harnesses

MAPLE skills published to registries

• SKILL.md at repo root: npx skills add kinncj/maple installs the full skill set
• skills/ dir: individual skills installable as npx skills add kinncj/maple@pipeline-runner, @tdd-workflow, @github-cli, etc.
• skills.sh indexes via install telemetry; skillsmp.com auto-scrapes GitHub — both will discover MAPLE automatically

Tests

• 17 Go unit tests for all output parsing functions (parseSkillsOutput, parseInstalledJSON, parseInstalledText, stripANSI)
• New tests/template/test_skills_sh.sh integration suite: live skills.sh API, install verification, publishing structure checks — 9 pass, CI-safe (network tests skip on CI=true)

v4.10.14 — taffy approval auto-resumes agent

Pressing [a] in the pipeline overlay now sends a continue keystroke to the agent's pane so the paused taffy resumes without the user having to switch windows and type it manually.

Mechanism per environment:

• Outer tmux: captures the new pane id on launch, tmux send-keys -t <pane> on approve
• Outer zellij: launches with a named tab, focuses + writes on approve
• No multiplexer but tmux installed: transparently wraps the launch in tmux new-session -d -s maple-<harness> and attaches from the spawned terminal, so approve still has a target
• No multiplexer, no tmux: file-deletion polling only (previous behaviour)

Works for claude, opencode, and copilot uniformly. Skill poll also bounded with timeout 540 and instructed to retry so approvals that take longer than the Bash tool's 10-minute cap no longer go silently ignored.

v4.10.13 — fix copilot launch crash

Fixes [x] quick prompt (and L launcher) crashing copilot. The copilot CLI does not accept a positional prompt — it errored out and closed the terminal immediately. Now uses -i/--interactive <prompt> so the prompt is executed in an interactive session.

v4.10.12 — quick prompt always shows harness picker

Fixes [x] quick prompt silently auto-selecting the first harness with a pinned session, which skipped the picker and failed silently when the stored session ID was stale. Picker is now always shown for any harness (claude, opencode, copilot) and annotates each with its pinned session so you can see what will resume.

v4.10.11 — rtk global hook wiring

What's new

• maple init / maple update: RTK is now wired globally with rtk init -g --auto-patch
  • Previously used rtk init (local, no-op without --auto-patch) — hook was never actually added to settings.json
  • -g adds the hook to ~/.claude/settings.json so it works across all projects
  • Also wires the OpenCode plugin with --opencode
  • Confirms hook status via rtk init --show after wiring
  • Idempotent — safe to re-run on maple update
• maple rtk-audit now shows three things in sequence:
  1. Hook wiring status (rtk init --show)
  2. Token savings analytics (rtk gain) — works without audit log
  3. Audit log (rtk hook-audit) — populated when sessions run with RTK_HOOK_AUDIT=1

v4.10.10 — rtk hook-audit wiring

What's new

• maple init now runs rtk hook-audit after rtk init to verify hooks are correctly wired — warns if they're not
• maple rtk-audit — new subcommand, direct alias for rtk hook-audit
• maple resume-session — sets RTK_HOOK_AUDIT=1 in the child process environment when rtk is on PATH
• All TUI-spawned harnesses ([x] quick launch, [L] launcher, [o] open session) — prefixed with env RTK_HOOK_AUDIT=1 so hook activity is logged automatically

v4.10.9 — [x] TAFFY-first picker with [t] toggle

What's new

• [x] quick launch defaults to TAFFY — opens the taffy workflow picker instead of skills/agents
• [t] toggles between modes — switch between TAFFY workflows and the skills/agents list; search resets and stays mode-aware
• Taffy badge shows stage count — [taffy 6 stages] so you know what you're launching before you select it
• Taffy items launch via /pipeline-runner <name> — skills/agents still launch as /<name>
• loadTaffyItems() reads .claude/taffy/*.yaml, parsing name, description, tags, and stage count