This prompt is a starting point for operating CTX with a new tool or integration.
Fill in the bracketed fields before use.
[Describe the primary objective in one clear sentence.]
- Includes: [what the agent should cover]
- Excludes: [what the agent must not touch]
- CTX repository: [local path]
- Cognitive branch: [branch name]
- Environment/stack: [runtime, SDK, dependencies]
- Constraints: [access, credentials, permissions]
The agent must work from CTX as the primary source:
ctx statusctx graph summaryctx logctx auditctx next
If ctx next returns no candidates, record the gap as a task before proceeding.
Do not edit .ctx manually unless there is a real operational block.
The agent must bind its work to CTX as the system of record:
- Always derive the next step from
ctx nextor explicitly recorded gaps. - Record every meaningful outcome as
evidence,decision, orconclusion. - Close a task with a cognitive commit before any Git commit.
- Do not invent work outside the CTX graph without adding the task/hypothesis first.
Describe the tool and how it integrates with CTX:
- Tool: [name]
- Expected input: [format]
- Expected output: [format]
- Evidence produced: [type]
- Limits or risks: [operational]
- Identify the active goal and task from CTX.
- Create a hypothesis if the justification is missing.
- Execute the smallest block that produces real evidence.
- Record evidence.
- Record a decision if direction changes.
- Close a conclusion.
ctx commit -m "<result>"- Git commit only after the cognitive commit.
- [expected results]
- [artifacts / files]
- [required evidence]
- The task is
Done. - Evidence and conclusion are linked.
- A cognitive commit exists.
- The Git commit reflects the real change.
Ask only if an external decision or credential is missing. If there is no real block, continue without asking for confirmation.
Distribution note:
- the reusable agent-link fragment is shipped in
distribution/agent-link/CTX_AGENT_LINK_PROMPT.txt