# Agent Install Prompt

`````txt
Summary of what we are doing:
- ask a short Greplica setup questionnaire before installing anything
- install greplica with the selected guidance and memory-update mode
- create initial memory from the codebase
- optionally analyze previous sessions

Before installing, ask me this short Greplica setup questionnaire. If this agent runtime supports native multiple-choice question UI, use it. Otherwise ask the same questions as plain multiple choice and wait for my answers. Do not run npm install, greplica install, bootstrap, or transcript search until I answer.

Question 1:
How should Greplica help agents during future sessions in this repo?

Options:
1. Guidance + auto-save (Recommended)
   Install hooks that remind agents to run `greplica graph context` before broad exploration, and automatically try to save useful memory after enough session activity.
2. Guidance only
   Install hooks only for in-session guidance. Agents get reminded to use `greplica graph context`, but Greplica will not automatically save memory after sessions.
3. No hooks
   Do not install hooks. Greplica will install normally, but future agents will only use it if you manually add the guidance snippet shown at the end.

Question 2:
Should I inspect recent prior sessions for this repo to seed useful Greplica memory?

Options:
1. Yes, recent sessions (Recommended)
   Find 1-3 recent same-repo sessions, show the selected transcripts, then store durable decisions, gotchas, and workflow context.
2. No, fresh start
   Only create baseline memory from current repo files and skip transcript search.

Map the first answer to install flags exactly:
- Guidance + auto-save: `--hooks enabled --auto-memory enabled`
- Guidance only: `--hooks enabled --auto-memory disabled`
- No hooks: `--hooks disabled --auto-memory disabled`

Then run:

```bash
npm install -g greplica
greplica install --platform <codex|claude|copilot|cursor|opencode|openhands|factory-droid|antigravity> --embedding local <mapped-hook-and-memory-flags>
```

Use the platform matching this agent. Do not manually copy skills. After installation, do not echo the full installer output or repeat its next steps.

Then bootstrap shallow memory for this repo:
- Prefer using the `greplica-bootstrap` skill.
- If the skill is not visible until restart, read the installed `greplica-bootstrap/SKILL.md` file and follow it directly.
- Create, validate, and apply the bootstrap proposal.
- Keep bootstrap output for the final answer to one line: `Greplica is installed and baseline memory was applied.`

If I chose "Yes, recent sessions", analyze prior sessions:
- Find recent prior sessions for this same repo and platform, preferring work from the last 1-2 days.
- Candidate locations: Codex `~/.codex/sessions/**/*.jsonl`; Claude Code `~/.claude/projects/**/*.jsonl`; GitHub Copilot CLI paths from `Stop` hook `transcript_path` values or `$COPILOT_HOME/session-state`; OpenCode `~/.local/share/opencode/storage/session/*.json` with messages under `storage/message/<sessionId>/`.
- Do not require transcript metadata `cwd` to equal the current checkout path. Users may use worktrees, renamed folders, or multiple checkouts of the same repo.
- Treat a transcript as same-repo when its metadata `cwd` is the current path, or when that `cwd` still exists and Git reports the same `remote.origin.url` or same normalized repo identity as the current repo. If the old path no longer exists, use transcript cwd text, repo name, branch, and recent session content as weaker matching evidence.
- Select 1-3 transcripts. Use one if there is a large high-signal session, two by default when multiple sessions are useful, and three only when sessions are smaller or cover distinct work.
- Show me the selected transcripts before bundling them: title if available, date/time, path, size/turn count if available, and why each matched this repo.
- Do not ask for confirmation. Continue with a temporary bundle path:

```bash
greplica transcript bundle --platform <codex-or-claude-or-copilot-or-opencode> --file <path-1> [--file <path-2>] [--file <path-3>] --out <greplica-transcript-backfill.md>
```

- Then use the `greplica-fast-session-bootstrap` skill on `<greplica-transcript-backfill.md>` and include its final value summary naturally in the final answer.

If I chose "No, fresh start", skip transcript search entirely.

If I chose "No hooks", do not edit `AGENTS.md`, `CLAUDE.md`, or any other agent instruction file. Instead, include this manual guidance snippet in the final answer and say I can add it to my agent instruction file if I want future sessions to use Greplica without hooks:

```md
Greplica hook guidance: greplica is a repo-memory search tool for finding relevant architecture, decisions, flows, and code anchors. Before broad manual exploration in this repository, run greplica graph context "<question>" with a focused natural-language query. When Greplica provides useful context, mention that you used it and briefly say what it helped with.
```

Final answer rules:
- Write like you are updating a human, not filling a template.
- Start by saying Greplica is installed and baseline memory is ready.
- Mention the selected guidance/memory-update mode: Guidance + auto-save, Guidance only, or No hooks.
- If transcript backfill ran, include the `greplica-fast-session-bootstrap` final value summary naturally.
- If transcript backfill was skipped, say it was skipped because I chose a fresh start.
- If hooks were installed, end with a short note that hooks and installed skills might need a restart or trust approval.
- If hooks were not installed, include the manual guidance snippet and do not tell me to accept hooks.
- Do not include installer output, selected transcript recap, proposal paths, apply counts, command lists, bundle paths, or a long usage guide unless I ask.
`````