Skip to content
Back to skills

Think

ASecurity

Emit a structured reasoning trace and persist it to substrate as a causal record for future /recall to walk.

  • 10 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added September 3, 2026
ai-agentsbash

Works with

  • mcp

Security analysis

A100/100

Scanned September 3, 2026

npx -y skills add xgre1/troth --skill think --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Think?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Think
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/xgre1-think/badge)](https://www.skillsdirectory.com/skills/xgre1-think)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: think
description: Emit a structured reasoning trace and persist it to substrate as a causal record for future /recall to walk.
argument-hint: [topic to reason about]
allowed-tools: [engram_record, engram_search, decision_record]
auto-persist: {"scope":"reasoning","salience":1}
---

User wants you to think out loud about:

> $ARGUMENTS

Reasoning protocol:

1. First, `engram_search({ query: "$ARGUMENTS", k: 4 })` to load what substrate already believes on this topic. Do not duplicate prior reasoning.
2. Reason in 3–5 bullet points: known facts → hypotheses → implication → recommended next action. Cite engram statements verbatim where they ground the reasoning.
3. Persist what the reasoning produced — two roads, pick by what it IS:
   - **It produced a reusable strategy** (a way to approach this class of
     problem again): `decision_record({ strategy, trigger, steps, contrast?,
     example?, provenance: { model: "<your model id>", verdict: "unverified" } })`.
     The trigger line is the retrieval key — write the SITUATION SHAPE it
     applies to, not the topic name. Include the contrast (mistake → why →
     correct) whenever the reasoning rejected a tempting wrong road — it is
     the highest-value field. Set verdict above `unverified` ONLY for what
     was actually verified: `test_passed` needs a test that ran, `operator_confirmed`
     needs the operator's word.
   - **It is a conclusion but not a strategy** (a fact, an assessment):
     `engram_record({ statement: "<one-sentence synthesis>", salience: 1, scope: "reasoning" })`.
4. Reply with the trace + the record id so the user can audit.

This is the canonical entry point for substrate-as-reasoning-history. The /think → engram → /recall → /think chain is what lets troth improve across sessions without re-thinking from scratch.

## Tool routing (both topologies)
The substrate tools this skill uses may be DIRECT in your tool list (names like
`troth_engram_record`, `troth_recall`, `troth_dialogue_recent`) OR behind the
troth-router gateway (app installs wire only: troth-router, troth-bash,
troth-cache, troth-hashline). If a named tool is NOT in your tool list, do NOT
conclude the substrate is down and do NOT fall back to file-based memory.
Route it through the router instead:
  1. `mcp_list({server: "troth-substrate"})` (or `"troth-memory"`) to see names.
  2. `mcp_call({server: "troth-substrate", tool: "<same troth_* name>", args: {...same args...}})`.
Substrate lives on server `troth-substrate` (engrams, recall, dialogue, slash);
mind/actions on `troth-memory`. Same tools, same args, one hop through mcp_call.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.