FUSE entry cache may not reflect the new session directory immediately
after the write to s/new returns. Poll until it appears, mirroring the
approach in s/sh.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Bash's printf builtin issues one write() syscall per call; the server
processes each write independently, so the first write (name= only)
would fail the cwd-required check. Build the full KV string first,
then write once.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Batch jobs were a front-end concern. All one-shot query workflows now
compose session primitives via s/b (engine), s/bfg (foreground), and
s/bbg (background). Parallel execution is a shell loop.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
tui/session: IsIdle reads spec and parses state= key.
scripts/s/sh, sh.rc: add spec_get helper; is_idle, /cwd, /model,
/agent, /backend read via spec_get; writes go to spec as key=value.
el/ellie.el: ellie--state reads spec with string-match on state=.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Option C implementation: single namespace with composable file tables.
Batch jobs are sessions with a reduced file set — the directory
contents declare the capability.
- core: SessionStore absorbs BatchStore; s/new detects batch format
- 9p: /b/ removed entirely; all routing through /s/
- scripts/b/* moved to scripts/s/*; paths updated
- subagent_spawn, optimize: /b/q -> /s/q
- scripts/b/ removed
Simplify the script to fix the timing/display problem where LLM responses
weren't shown in real-time and user typing wasn't echoed properly.
The fix:
1. Block on statewait file until agent leaves idle state
2. Stream chat output in loop while agent is not idle
3. Use simple tail -c +offset to read new chat content
4. Remove complex buffering logic and background processes
This resolves the issue where assistant responses would appear one prompt
late and user typing wasn't visible during LLM response generation.
b/job now blocks on statewait instead of polling; b/cleanup reads state.
Update all documentation references. Bump 9p submodule.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove Go template syntax from all prompt files (converted to envsubst
variables in prior commits). Agent agentSpawn hooks now call prime for
each prompt file explicitly, replacing the automatic Go-based directory
scan. coding agent gets 08_tool-elevate; default does not.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
prime <name> prints $OLLIE/p/<name>[.md] to stdout, which the agentSpawn
hook captures as context. Allows agent prompt composition to be driven
entirely from hook config rather than built-in Go code.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Submodules have .git as a file, not a directory. -d was stopping the
walk immediately in submodule directories.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Walk up from CWD only while .git exists at each level. Stops at the
first directory without .git, preventing false positives outside the repo.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The bd wrapper walks up from CWD to find a .beads directory and sets
BEADS_DIR, so bd works from any subdirectory of a monorepo without the
model needing to know about it.
The execute server now always prepends $OLLIE/x/ to PATH for tool
subprocesses, so wrappers placed there shadow system binaries transparently.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace tools/elevate_superpowers with scripts/x/elevate — a single
self-contained plugin that bootstraps into a superpowers session and
runs the IPC. mkfile installs it to ~/.config/ollie/scripts/x/.
Update 08_tool-elevate.md to remove backend-specific language.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- s/sh: add -e VAR1,VAR2 flag to inject named vars from frontend env
into session spec as env.NAME=VALUE lines
- s/sh: remove $MOUNT/env export loop (global env file removed)
- Bump core and 9p submodules
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add explicit handlers for all session ctl commands (/stop, /kill,
/compact, /clear, /rn, /cwd) and session file r/w commands (/agent,
/model, /backend) and read-only queries (/models, /backends, /usage,
/ctxsz). Ctl commands now continue past the drain loop. /rn updates
local $id/$S after rename. Unknown slash commands print an error
instead of writing garbage to ctl.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Generates N candidate optimized prompts in parallel, then judges and
returns the best. Accepts -cb/-cm for candidate model and -jb/-jm for
judge model, defaulting to OLLIE_BACKEND/OLLIE_MODEL for both stages.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Install scripts to ~/.config/ollie/scripts/ (not ~/bin/) so they are
served via the 9p mount without PATH changes. Add thin wrappers oq-q
(s/q, foreground one-shot) and oq-schedule (s/sched, background with
b/ path output). Switch osh and oq from OLLIE_9MOUNT to OLLIE.
Update mkfile, docs, and submodule pointer.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
-bg submits the job and exits immediately, printing session id(s) for
later retrieval via b/<id>/result; implicitly enables -keep.
A trailing newline is printed on all exit paths via an EXIT trap.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Brace group exit code was the last [ -n ] test, which returns 1 when
the variable is empty, triggering || die even on success. Adding true
as the final statement ensures a clean exit.
Replace write to defunct /ctl with s/new spec format. Add -name flag
so named sessions skip the post-write poll. Fix /kill to write to
s/{id}/ctl instead of the removed top-level ctl file.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Levi Neely