菜单
Consolidate a contributor's L1 atoms into an L3 persona across 11 dimensions. Triggers when the user says "build contributor persona", "consolidate <user>", or after contrib-ingest completes. This ORGANIZES existing atoms — for creating atoms use contrib-ingest.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use the tmem CLI to inspect and manage the local memory store — record/vector counts, keyword/hybrid search, view persona, list scenes, read one full scene block (tmem scene <name>), recent changes, sync vectors, configure thresholds. Also covers CROSS-PROJECT exploration — search every project store at once (tmem search <q> --all), list all memory stores (tmem projects), and collapse legacy cwd-keyed fragment stores into their project root (tmem migrate-fragments). Trigger when the user asks "how many memories do I have", "show my persona", "what scenes exist", "search memories for X", "search across all projects", "what projects/stores do I have", "my memory is fragmented", "open that scene", "recent memory changes", or when you need to check memory state before/after an operation. Do NOT use for extracting or consolidating memories — those have dedicated skills.
Consolidate L1 memory atoms into L2 scene blocks and L3 persona. Triggers when the user says "consolidate memories", "build persona", "update persona", "update scenes", "organize memories", or after memory-seed completes. Also triggers via asyncRewake pipeline after N conversation turns. This skill is about ORGANIZING existing memories into higher structures — for creating memories from transcripts use memory-seed instead.
Extract L1 memory atoms from Claude Code conversation history. Triggers when the user says "seed memories", "extract memories", "backfill memory", "remember my history", "learn from past conversations", "what do you know about me", or after /memory-init when conversation history exists. Also use when asyncRewake pipeline flags pending sessions. This skill is about CREATING new memories from transcripts — for inspecting existing memories use tmem-cli instead.
One-shot entry point for Contributor Intelligence — the user just drops a GitHub link or names an engineer and you take it from there. Use this whenever someone pastes a GitHub profile/repo URL (github.com/<user> or github.com/<user>/<repo>) or a handle and asks to "profile", "analyze how X works", "learn from X", "study this engineer", "phân tích người này", "học từ người này", or asks "how do I use /contrib". Drives the whole pipeline end-to-end (add → ingest → build → playbook), or guides the user through it if they'd rather drive. This is the orchestrator — the per-phase skills (contrib-ingest, contrib-consolidate, contrib-synthesize) do the actual work; do NOT use this one to inspect or synthesize an already-built persona (use contrib-synthesize) or for the self-memory feature.
Extract L1 contributor atoms from a GitHub subject's raw activity. Triggers after "tmem contrib raw <id>" or when the user says "ingest contributor", "profile <user>", "analyze how <user> works". This CREATES atoms from raw GitHub events — for synthesis use contrib-synthesize.
Use this skill when someone wants to LEARN from engineers they've already profiled — extracting actionable lessons from contributor personas. Trigger on intents like: "what do the top SWEs / engineers / contributors I've profiled have in common", "extract their shared / common capabilities", "turn a contributor's profile into copyable habits, heuristics, or a learnable playbook", "what habits should I copy from <person>", or "compare me / my style to <role-model> dimension-by-dimension and find my gap". Also covers surfacing exemplar teaching quotes across profiled engineers. Operates on existing personas (built via contrib-consolidate); cross-engineer comparison or common-capability synthesis needs ≥2 subjects. Do NOT use for: building, ingesting, or inspecting a single persona's raw contents; commit/trajectory stats; summarizing documents; recalling what you know about the user; or playbooks unrelated to engineer profiles (e.g. team incident-response runbooks).
| name | contrib-consolidate |
| description | Consolidate a contributor's L1 atoms into an L3 persona across 11 dimensions. Triggers when the user says "build contributor persona", "consolidate <user>", or after contrib-ingest completes. This ORGANIZES existing atoms — for creating atoms use contrib-ingest. |
Group one subject's L1 atoms into themes (L2 scenes, conceptual) and write a single L3 persona summarising all 11 dimensions with evidence.
tmem contrib atoms <subject-id>
Read references/persona-guide.md first — it covers merging atoms (not
listing), resolving conflicting atoms, weighting by evidence, when to mark
"insufficient data", and the quality bar (a persona should predict how the
subject tackles a new task).
For each of the 11 dimensions (idea, plan, solve, craft, comms, mentor, conflict, scope, ownership, execution), synthesise the atoms in that dimension
into 1–3 sentences. Carry the strongest evidence links into the text. If a
dimension has no atoms, set it to "insufficient data".
Collect 3–6 notable_traits — distinctive things that don't fit a fixed
dimension (e.g. "writes prose-quality commit bodies", "prefers small composable
modules").
tmem contrib upsert-persona --json '{
"subject_id": "<id>",
"summary": "<2-3 sentence overview of how this engineer works>",
"dimensions": {
"idea": "...", "plan": "...", "solve": "...", "craft": "...",
"comms": "...", "mentor": "...", "conflict": "...",
"scope": "...", "ownership": "...", "execution": "..."
},
"notable_traits": ["...", "..."],
"updated_time": "<ISO timestamp>"
}'
Print the persona (tmem contrib persona <id>) and tell the user which
dimensions are well-evidenced vs "insufficient data".