| name | orient-digest |
| description | Weekly personal orientation digest for Stephanie over the code-intel-digest mirror. Use when she wants to catch up on the last ~7 days of AI/dev/research news for herself (not to share). Signal-dense, terse, assumes Stephanie-level context. For share-ready output, use distill-digest instead. Requires the code-intel-copilot MCP server. |
| allowed-tools | mcp__code-intel-copilot__search_items, mcp__code-intel-copilot__semantic_search_items, mcp__code-intel-copilot__get_item, mcp__code-intel-copilot__aggregate_items, mcp__code-intel-copilot__mirror_status, Read, Write, AskUserQuestion |
Orient Digest
Personal situational-awareness digest. Audience: Stephanie only. She already knows the landscape — this is catch-up, not onboarding. Skip explanations of things she obviously knows (what Cursor is, what MCP is, what SCIP stands for).
Always do first
- Call
mirror_status exactly once and check dbMode. In direct mode the server reads the live production database (real-time) — note "Data: live production (real-time)" at the top. In mirror mode, if staleMinutes > 90 warn and ask before continuing; otherwise note freshness in one line ("Data through ").
- Ask for a lens/prioritization once via
AskUserQuestion — offer 3-4 options drawn from the default watchlist (below) plus "let the data pick". Skip this step if she already specified one in her prompt.
Defaults
- Time window: rolling 7 days (
since = today - 7d as ISO date). Only deviate if she asks.
- Format: themed sections. For each theme: 1-2 sentence synthesis, then 2-5 items as a tight bulleted list (
- [title](url) — one clause why-it-matters). 3-6 themes total.
- Length: short. ~300-500 words total. She's orienting, not studying.
- Output: inline (no file write unless she asks).
Watchlist (boost when ranking)
- Competitive AI-coding-tool market — Cursor, Copilot, Claude Code, Codex, Windsurf, Augment, CodeRabbit, Cognition/Devin, GitLab Duo. Launches, market-share data, pricing, narrative shifts.
- Enterprise AI rollouts & case studies — named-company posts about internal platforms, adoption numbers, ROI claims (Cloudflare, Meta, Intercom, Atlassian, etc.).
- Code context / codebase understanding / SCIP / MCP — the core Sourcegraph thesis. Anything about agent context, repo-level tools, long-context code navigation, cross-repo intelligence.
De-prioritize
- Reddit / community venting posts. Keep only if they show a clear buyer-sentiment shift (e.g., multiple threads on the same pain point, or someone canceling a subscription and explaining why in detail). Skip anecdote-only threads.
Core principles
- Cite URL inline for every item mentioned. No bare titles.
- Don't overclaim. "3 items discussing X" ≠ "the industry is doing X".
- Respect
avgFinalScore from aggregate_items — higher = more useful-to-Stephanie signal.
- Dates: prefer
since=YYYY-MM-DD as ISO (7 days before today).
- No enumeration. Summarize, group, synthesize.
- No hand-holding. No "what is MCP?" explanations. No "this means Sourcegraph should consider..." framings. Just the signal.
Recipe
mirror_status.- Ask for lens unless prompt specifies one.
- Gather candidates:
search_items(since=<7d>, limit=50) — general sweep.aggregate_items(group_by='source', since=<7d>, limit=15) — see who's active.- For each watchlist pillar she prioritized, run a targeted
semantic_search_items query (limit 10-15 each).
- De-dup by
id. Drop anything matching the de-prioritize list unless it's flagged as buyer-sentiment shift.
- Cluster into 3-6 themes by what's actually strong this week (not a forced template).
- For each theme: 1-sentence synthesis + 2-5 items.
- End with "What I'd watch next week" — 1-3 specific signals to check on.
Guardrails
- Don't write Sourcegraph-GTM angles here — that's distill-digest's job. This is for Stephanie orienting.
- If a theme only has 1 item, either promote it to a standalone callout or drop it. No themes-of-one.
- Sourcegraph and Amp are separate companies; treat Amp as a competitor, not a Sourcegraph product.
- If 0 items match the lens she picked, say so directly and offer a fallback lens.
