| name | agora-code |
| description | Use agora-code memory tools — inject session context, learn findings, recall past work, checkpoint progress, and summarize large files |
STOP — before you do anything else:
Run agora-code summarize <file> before reading ANY file over ~50 lines.
Do NOT use Read, shell commands (cat, head, tail, etc.), or an Explore subagent on a large file without summarizing first.
This is not optional. The Explore subagent is mechanically blocked. Shell commands bypass hooks — don't use them to read files.
agora-code gives you persistent memory across sessions. Hooks handle most things automatically, but you must follow the rules below — they are not optional.
Your rules (always follow these)
- Before reading any file over ~50 lines — run
agora-code summarize <file> first. Do not use the Read tool or an Explore subagent on a large file without summarizing first. This is mandatory, not optional.
- At session start — run
agora-code inject to load prior context (checkpoints, learnings, git state, symbol index).
- When done with a task — run
agora-code complete --summary "..." to archive the session.
- Always run
agora-code status -p (not status) to see per-project stats.
- Do not ask the user to set a goal — infer it from what they're working on.
What the hooks handle automatically (no action needed)
| Hook | Event | Does |
|---|
pre-agent.sh | PreToolUse(Agent) | Blocks Explore subagent — hooks don't fire inside subagents |
pre-read.sh | PreToolUse(Read) | Intercepts large files — auto-summarizes before Claude reads |
on-read.sh | PostToolUse(Read) | Indexes symbols + code blocks into DB |
on-grep.sh | PostToolUse(Grep) | Indexes files matched by grep into DB |
on-edit.sh | PostToolUse(Write/Edit) | Re-indexes symbols, tracks diff |
on-bash.sh | PostToolUse(Bash) | Tags committed files with SHA on git commit |
on-prompt.sh | UserPromptSubmit | Auto-sets goal, recalls relevant learnings |
on-stop.sh | Stop | Parses transcript → structured checkpoint |
PostCompact re-injects context automatically after context compaction.
Session lifecycle
SessionStart → agora-code inject # load prior context
Working → agora-code summarize <file> FIRST # ← MANDATORY before any large file
Step done → agora-code checkpoint --goal "..." # optional mid-task save
All done → agora-code complete --summary "..." # archive session
Manual commands reference
| Command | When to use |
|---|
agora-code inject | Load prior session context |
agora-code summarize <file> | Before reading any file over ~50 lines |
agora-code learn "<text>" | Force-save a specific finding right now |
agora-code recall "<query>" | Search past findings for a topic |
agora-code checkpoint --goal "..." | Save progress mid-task |
agora-code status -p | Check session and DB stats for this project |
agora-code complete --summary "..." | Archive session when done |
inject output format
agora-code inject outputs ~300 tokens of structured context:
LAST CHECKPOINT
goal / decisions / next_steps / blockers / files / branch + commit
LEARNINGS
recent findings tagged by type (decision → / blocker ! / next » / finding ·)
GIT LOG
last 6 commits
UNCOMMITTED
dirty files
SYMBOL INDEX
function:line for dirty files (avoids re-reading them)
