rik

Added

• Discover agent skills from ~/.agents/skills, ~/.codex/skills, and ~/.claude/skills, preferring the first location that defines a given skill name
• Add a skill tool that loads a skill's instructions and its bundled reference files, and advertise the available skills to the agent for both tasks and questions
• Preload skills into the agent preamble with --skills, failing with the full list of available skills when a name is unknown
• Close a multiline marker with a delimiter trailing its last content line (rik: [[ … third line ]]) instead of requiring the delimiter on a line of its own
• Remember each completed turn — tasks as a model-written what/why/how summary, questions as the question and its answer — and replay the summaries to later turns as a History block. Budgeted with memory-tokens / --memory-tokens (default 32000, 0 disables); on overflow the oldest two thirds are merged into one summary and the newest third follows it
• Add a recall tool so the agent can read a remembered turn in full — request, reasoning, output, or diff — by id. Diffs and raw turn text are held outside the prompt, so the budget covers many more turns
• Stop a running task as soon as the user edits the file being worked on, leaving both rik's partial work and the user's edit in place (no revert)

Changed

• Update to rig-core 0.40, toml 1, and shlex 2, migrating tools to the split description() / parameters() trait methods and the unified PromptResponse