priivacy-ai/ad-hoc-profile-load

ad-hoc-profile-load

Load an agent profile interactively to adopt a specific role for the current session. This skill is for ad-hoc use outside the mission runtime loop — when a user wants an agent to behave as a particular role without starting a formal mission.

---

When to Use This Skill

Use this when the user asks you to:

• Act as a specific agent role (architect, reviewer, implementer, etc.)
• Load a profile for an interactive session
• Adopt role-scoped boundaries and governance context
• Switch roles mid-conversation

Do NOT use this when:

• A mission is running and spec-kitty next is driving the loop — the runtime assigns profiles automatically via DDR-011 matching
• The user wants to create a new profile — use the charter synthesize workflow (spec-kitty charter synthesize) or author the profile YAML directly
• The user wants to modify an existing profile — edit the YAML directly

---

Step 1: Resolve the Profile

If the user names a profile directly, load it. If they describe a role or task, resolve the best match.

Inspect by explicit ID — read the full resolved definition before adopting it:

spec-kitty agent profile show <profile-id>
# Add --all to inspect a non-activated abstract base profile:
spec-kitty agent profile show <profile-id> --all

from doctrine.agent_profiles import AgentProfileRepository

repo = AgentProfileRepository(project_dir=project_agents_dir)
profile = repo.resolve_profile("architect")

By task context (when the user describes what they want to do):

from doctrine.agent_profiles.profile import TaskContext

context = TaskContext(
    languages=["python"],
    frameworks=["fastapi", "<project-test-runner>"],
    file_patterns=["src/**/*.py"],
    domain_keywords=["architecture", "design"],
)

profile = repo.find_best_match(context)

Discovery (when the user is unsure which profile to use):

spec-kitty agent profile list

One-shot invocation (when the user wants a profile-governed answer rather than to adopt the role for the session): route the request through the canonical invoke surfaces instead of loading the profile manually:

# Ask a named profile to handle a specific request:
spec-kitty ask <profile-id> "<request>"

# Or let the router pick a profile for the request:
spec-kitty advise "<request>" --profile <profile-id>

---

Step 2: Apply the Profile

Once resolved, adopt the profile by internalizing three things:

Identity

Read initialization_declaration — this is your startup persona statement. Acknowledge it at the beginning of the session.

print(profile.initialization_declaration)
# "I am Architect Alphonso. I design scalable, maintainable system
#  architectures using established design patterns and principles..."

Boundaries

Read specialization — this defines your scope:

• primary_focus — what you actively do
• secondary_awareness — what you consider but don't own
• avoidance_boundary — what you must not do

Before taking any action, check whether it falls within your boundaries. If the user asks you to do something in the avoidance boundary, acknowledge the request and explain which role would handle it instead (using collaboration.handoff_to).

Governance Scope

The profile's context_sources declares which doctrine layers and specific directives are relevant to this role. Load only those:

from doctrine.service import DoctrineService

service = DoctrineService(shipped_root, project_root)

# Load directives referenced by this profile
for ref in profile.directive_references:
    directive = service.directives.get(f"DIRECTIVE_{ref.code}")
    # Apply this directive's constraints to your behavior

Do NOT load the full doctrine catalog. The profile scopes what matters.

---

Step 3: Scope Governance Context

After adopting the profile, load charter context scoped to the action the user wants to perform:

spec-kitty charter context --action implement --json

If the user hasn't named an action, infer it from the profile's canonical_verbs:

| Profile | Canonical verbs | Default action | |---|---|---| | architect | design, evaluate, decide, model, specify | specify | | planner | plan, prioritize, decompose, schedule | plan | | implementer | implement, fix, refactor, test, debug | implement | | reviewer | review, approve, reject, assess | review | | researcher | research, investigate, evaluate, synthesize | specify | | curator | curate, validate, update, reconcile | review | | designer | design, prototype, sketch, iterate | specify |

---

Step 4: Maintain Role Throughout the Session

Respect Handoffs

When work falls outside your boundaries, name the appropriate role:

"This requires implementation work. That's in Implementer Ivan's scope —
I can hand off my architectural notes for them to execute."

The profile's collaboration section defines:

• handoff_to — roles you delegate work to
• handoff_from — roles that delegate to you
• works_with — roles you collaborate with in parallel

Pull Doctrine On Demand

When you need guidance mid-session, pull specific tactics or directives relevant to your current task — don't reload the full context:

# Need guidance on a design decision?
tactic = service.tactics.get("adr-drafting-workflow")

# Need to check a quality gate?
directive = service.directives.get("DIRECTIVE_030")

Mode Selection

The profile's mode_defaults lists the working modes this role supports. If the user's request maps to a specific mode, acknowledge it:

for mode in profile.mode_defaults:
    # mode.mode → "analysis", "design", "review", etc.
    # mode.description → what this mode is for
    # mode.use_case → when to activate it
    pass

---

Step 5: Tool Context Persistence (Optional)

To persist the profile for the current tool so it loads automatically on next session, start an ad-hoc specialist session via the slash command:

/spec-kitty.profile-context <profile-id>

This anchors the named profile as the active advisory context for the tool so it is available to the agent on subsequent turns. For a single profile-governed answer without persisting context, use spec-kitty ask or spec-kitty advise (see Step 1).

---

Quick Reference

# List profiles
spec-kitty agent profile list

# Inspect a profile (add --all for non-activated abstract base profiles)
spec-kitty agent profile show architect
spec-kitty agent profile show architect --all

# One-shot profile-governed invocation
spec-kitty ask architect "<request>"
spec-kitty advise "<request>" --profile architect

# Persist a profile as the tool's advisory context (slash command)
# /spec-kitty.profile-context architect

from doctrine.agent_profiles import AgentProfileRepository
from doctrine.service import DoctrineService

# Load profile
repo = AgentProfileRepository()
profile = repo.resolve_profile("architect")

# Read identity
profile.initialization_declaration
profile.specialization.primary_focus
profile.specialization.avoidance_boundary

# Load scoped doctrine
service = DoctrineService(shipped_root, project_root)
for ref in profile.directive_references:
    service.directives.get(f"DIRECTIVE_{ref.code}")

# Check boundaries before acting
if task_type in profile.specialization.avoidance_boundary:
    suggest_handoff(profile.collaboration.handoff_to)