How do I install the Context Trace skill?

Run npx skillmds@latest add ai-analyst-lab/context-trace in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Context Trace safe to use?

This skill has not completed SkillMD's automated safety review yet. Capability flags: docs only. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Context Trace?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Context Trace free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Context Trace?

ai-analyst-lab (@ai-analyst-lab) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Context Trace by ai-analyst-lab — Agent Skill

Name: Context Trace
Rating: 4.2 (5 reviews)
Author: ai-analyst-lab

Context Trace

Show which context the AI analyst would receive for a question, why each item was selected, how it is delivered, and which conflicts, stale items, or omissions need attention. Use when the user asks what context was loaded, why a definition did or did not reach an analysis, what the analyst knows about a question, or invokes /context-trace.

File contents

Context trace

Build an evidence record before explaining the context.

Read .knowledge/active.yaml and confirm an active dataset exists.
Run the deterministic manifest builder. Replace the question and output path with the user's values:
```
python -m helpers.knowledge.context_manifest \
  --project-root . \
  --question "THE EXACT QUESTION" \
  --output working/context-trace.json
```
When a live DuckDB path is available, add --duckdb PATH. This runs the schema guard and excludes relevant items whose table shape no longer matches the reviewed snapshot. For another warehouse, call guard_context() with the live connection and pass its result to build_context_manifest(quarantined=...).
Read working/context-trace.md and working/context-trace.json.
If blocking is true, stop the analysis and show the conflicting trusted definitions or relevant schema quarantine. Ask for an approved decision or re-verification.
If a selected item is stale or missing a review date, name it before using it.
Explain the result in four parts:
- content: what information was supplied;
- representation: metric, relationship, example, correction, or instruction;
- delivery: resident, selected, or compiled;
- governance: source, owner, status, review date, conflict, and omission.
State the limit plainly: the manifest proves what the selector supplied. It does not prove the worker used it or that it is correct.

When an analysis is available, ask the worker to return the context item IDs it cited and used. Reconcile those IDs and the SQL with reconcile_context_use() from helpers.knowledge.context_manifest. Distinguish supplied, cited, and applied. Never treat a citation alone as proof that the SQL followed the context.

---name: context-tracedescription: Show which context the AI analyst would receive for a question, why each item was selected, how it is delivered, and which conflicts, stale items, or omissions need attention. Use when the user asks what context was loaded, why a definition did or did not reach an analysis, what the analyst knows about a question, or invokes /context-trace.---# Context traceBuild an evidence record before explaining the context.1. Read `.knowledge/active.yaml` and confirm an active dataset exists.2. Run the deterministic manifest builder. Replace the question and output path with the user's values: ```bash python -m helpers.knowledge.context_manifest \ --project-root . \ --question "THE EXACT QUESTION" \ --output working/context-trace.json ``` When a live DuckDB path is available, add `--duckdb PATH`. This runs the schema guard and excludes relevant items whose table shape no longer matches the reviewed snapshot. For another warehouse, call `guard_context()` with the live connection and pass its result to `build_context_manifest(quarantined=...)`.3. Read `working/context-trace.md` and `working/context-trace.json`.4. If `blocking` is true, stop the analysis and show the conflicting trusted definitions or relevant schema quarantine. Ask for an approved decision or re-verification.5. If a selected item is stale or missing a review date, name it before using it.6. Explain the result in four parts: - content: what information was supplied; - representation: metric, relationship, example, correction, or instruction; - delivery: resident, selected, or compiled; - governance: source, owner, status, review date, conflict, and omission.7. State the limit plainly: the manifest proves what the selector supplied. It does not prove the worker used it or that it is correct.When an analysis is available, ask the worker to return the context item IDs it cited and used. Reconcile those IDs and the SQL with `reconcile_context_use()` from `helpers.knowledge.context_manifest`. Distinguish supplied, cited, and applied. Never treat a citation alone as proof that the SQL followed the context.

Frequently asked questions