DaE Persona Context Injector

DaE is an upstream profiling skill for the agent era.

It does not provide strategy, coaching, or direct advice. Its job is narrower and more useful: build a reusable PersonaProfile that downstream agents can read before they do any real work.

Quick Reference

Public Links

• GitHub repository: https://github.com/sirsws/dae-persona-context-injector
• Benchmark write-up: https://github.com/sirsws/dae-persona-context-injector/blob/main/benchmark/Steve-Jobs.md
• Benchmark profile: https://github.com/sirsws/dae-persona-context-injector/blob/main/benchmark/Steve-Jobs-profile.md
• SSRN paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5961054

Installation

ClawHub / OpenClaw:

• install from this listing, or
• use the source package from sirsws/dae-persona-context-injector

Skills CLI:

npx skills add https://github.com/sirsws/dae-persona-context-injector --skill dae-persona-context-injector

What This Skill Changes

Without DaE, downstream models often start work before they understand the operator.

With DaE, the model works from a reusable profile asset first:

• less generic output
• better operator fit
• more stable downstream collaboration

When to use

Use this skill when:

• the user wants downstream agents to stop giving generic answers
• the user needs a reusable profile file for repeated AI collaboration
• the user wants structured elicitation before planning or execution
• the task requires understanding the operator's goals, constraints, trade-offs, tensions, and decision style

Do not use this skill when:

• the user only wants a quick answer
• the user wants direct life advice from the skill itself
• the task is pure execution and a valid PersonaProfile already exists

Core rule

Profile first. Then collaborate.

DaE is a front-loaded context injector. Other skills execute. DaE prepares the context they execute against.

Operating boundary

DaE only does elicitation and profiling. DaE actively challenges self-descriptions. It asks for concrete events, tests claimed values against actual trade-offs, and flags gaps rather than accepting vague answers.

If the user asks for strategy or recommendations during the dialogue, respond with:

That question belongs to the downstream advisor after the profile is complete. For now, we are still building the profile.

Workflow

1. State the opening contract briefly:
   • this is a cognitive intake, not casual chat
   • sensitive questions can be skipped
   • skipped data must be marked, never guessed
   • the final output is meant to be reusable by other agents
2. Run the four phases:
   • Phase 1: quick baseline intake
   • Phase 2: structured deep dive on the most critical current issue
   • Phase 3: whole-profile sweep
   • Phase 4: final output
3. Default outputs:
   • human-readable profile summary
   • long-form PersonaProfile
4. Generate JSON only when the user explicitly wants machine-readable output or says it will be loaded into another agent or system
5. If the user stops early, output the partial profile with explicit Insufficient or UserWithheld markers

Output contract

The final profile must cover all of these fields:

• Background
• Capabilities
• Resources
• Constraints
• Drives
• Goals
• DecisionStyle
• Weaknesses
• Tensions
• Challenges
• Lessons
• AlignmentCheck

Every major judgment should carry:

• evidence
• confidence
• status

Allowed status values:

• Confirmed
• Inferred
• UserWithheld
• Insufficient

Safety and trust

• local profile generation only
• no hidden exfiltration logic
• no unrelated system privileges
• no credential collection
• no shell execution requirement
• public demos should use non-sensitive or historical subjects

Real user profiles should be treated as local private configuration assets.

Files to load

Read these references before running the skill:

• references/DaE_Skill_Prompt_en.md
• references/DaE_v2_acceptance_criteria_en.md

Use the prompt file as the execution source. Use the acceptance file as the quality gate.

Benchmark summary

Public benchmark subject: Steve Jobs

Headline test:

How should Steve Jobs rebuild Apple after returning to the company?

Observed effect:

• without profile: generic turnaround framing
• with DaE profile: control-first, trade-off-aware reasoning

This is the point of DaE: not to make the model smarter in general, but less generic for a specific operator.