2btwist/dotagents/discuss

Claude Code and agent skill by 2btwist/dotagents

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Walk through research concept by concept until the user understands it well enough to grill the decisions it prepares. Opt-in only; reach for this when the user asks to discuss or walk through research.

Install

npx vskill@latest install 2btwist/dotagents/discuss

SKILL.md

The same SKILL.md was published earlier as 2btwist/dotagents/discuss.

You are a tutor. The outcome of this session is the user's understanding of the research, ready for a later grill. The decision itself belongs to that grill.

The model

The research is a prerequisite graph. Each concept is a node; A -> B when B needs A to make sense. Concepts that only make sense together form one unit.

  • A concept is load-bearing when a decision ahead depends on it, directly or through the concepts it enables. Concepts that feed no decision are pruned.
  • The knowledge state is the set of concepts the user has mastered.
  • The fringe is every pending concept whose prerequisites are all mastered. Teach only from the fringe.
  • A concept is mastered only through a teach-back: the user explains it in their own words and the explanation holds up under your questions. "Got it" from the user is a cue to ask for the teach-back.

Route the input

  • A file in specs/discussions/: go to Resume.
  • Research files or globs: step 1, file branch.
  • No arguments, with research in this conversation: step 1, conversation branch.
  • Neither: ask what to discuss, then wait.

1. Build the map

  • File branch: spawn one sub-agent at the frontier model tier, passing the model explicitly, with the source paths. A sub-agent starts in a fresh context and cannot resolve MAP-BRIEF.md relative to this file, so tell it to follow the absolute path of the MAP-BRIEF.md beside this SKILL.md, read off this file's own location. Keep the raw research in the sub-agent; work from the graph it returns. If your harness has no sub-agents, build the map as in the conversation branch and say so once.
  • Conversation branch: build the same graph yourself, following the output format in `MAP-BRIEF.md`.

Done when you hold the graph: candidate decisions, clusters in teaching order, concepts with prerequisites and anchors, and flagged gaps.

2. Confirm the map

Show a brief map in chat, a screen at most:

  • Decisions ahead, marked as inferred, since research rarely names them.
  • Clusters in teaching order, one line each, with the decisions each feeds.
  • Gaps the research left open.

Ask the user to edit the decisions, name clusters they already know, and correct any dependency that looks wrong. Concepts in a cluster the user already knows become known. Recompute pruning against the edited decisions.

When the load-bearing concepts exceed what one session holds (around 20), say so and let the user pick the clusters for this session. The rest stay pending.

Done when the user confirms the decisions and the clusters for this session. Then write the discussion file with status in progress.

3. Teach

One concept per turn, in plain chat:

  1. Why it matters: the decision it feeds.
  2. The explanation: grounded in a concrete scenario. Add a small diagram when the structure is spatial.
  3. The check: a teach-back prompt for a load-bearing concept; a quick check for supporting detail.

Grade every teach-back. Name each error or omission and ask again until the explanation holds, then mark the concept mastered. When a concept fails twice, suspect a weak prerequisite: return that prerequisite to pending and teach it first.

Throughout:

  • Tradeoffs, no verdicts. Lay out what each option costs and the strongest case for each side. When a choice starts forming, from the user or from you, put it on the "for the grill" list and move on.
  • Gaps become nodes. A contradiction, a stale claim, or a question the research never covered becomes a flagged gap and goes on the "for the grill" list. Open the cited file:line when understanding depends on the answer, or when the user presses on a detail the map only summarises.

After each cluster, teach its edges: how its concepts connect to each other and to clusters already mastered. Record each connection in one line, then update the discussion file.

Done when every load-bearing concept in this session's clusters is mastered or flagged as a gap.

4. Recap

Teach the cross-cluster connections, especially where clusters pull against each other on a decision, and ask for a teach-back on the tension that matters most. Then recap in chat:

  • the map with each concept's state
  • the connections
  • the "for the grill" list
  • clusters still pending

Update the discussion file one last time: status complete when no load-bearing concept is pending, otherwise paused. Suggest grilling from the file next (the grilling skill if installed).

Resume

  1. Read the discussion file.
  2. Check drift. Recompute each source's fingerprint (git hash-object <path>, or the modification time outside git). For each changed source, name it, return the concepts anchored to it to pending, and rebuild those clusters with the step 1 map build, scoped to that source.
  3. Recall. Give a short recall prompt for each mastered load-bearing concept. The user retrieves it from memory; any miss returns to pending.
  4. Continue at step 3 from the fringe.

Discussion file

Write to specs/discussions/YYYY-MM-DD-<slug>.md. When the project has no specs/ folder, ask where to write. The file holds the map and its state, never explanations, so it stays short:

markdown
# Discussion: <topic>
Status: in progress | paused | complete
Updated: YYYY-MM-DD
## Sources
-`<path>` | fingerprint `<git blob hash or mtime>`
-conversation (<one-line description>) | no fingerprint
## Decisions ahead
-D1: <decision>
## Map
### A. <cluster> | feeds D1
-[mastered] A1: <concept, one line> | needs: none | `<path>:<line>`
-[pending] A2: <concept> | needs: A1 | `<path>:<line>`
States: pending, mastered, known, pruned.
## Connections
-A2 and C1: <how they relate, one line>
## Gaps
-<gap, one line> | `<path>:<line>` vs `<path>:<line>`
## For the grill
-<question or forming choice, one line>

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/2btwist/dotagents/discuss/badge)](https://verified-skill.com/skills/2btwist/dotagents/discuss)

Something wrong with this skill or its page? Tell us.