SoherAgent · 1 of 5 · Plan the idea
Proposed /plan concept · sanitized recreation
Scene 1 of 13 — Open the /plan source selector
Run one command, then choose where the starting idea should come from.
Proposed selector · not implemented runtime evidence
soheragent[concept] ›/plan on opens a source selector. It never requires a second terminal command to submit the idea.explanation
you@pi:~$/plan on
Scene 2 of 13 — Prune the session, then draft
The agent keeps valuable planning context and removes repetition, tangents, and raw conversational noise.
Read-only pruning · session unchanged · raw transcript not copied
User source recordedUse this session’s discussioninput source only
soheragent[plan-concept] $prune current-session context → synthesize proposed idea briefagent action
context/prune ✓ retained: problem · users · outcomes · constraints · decisions · open questions
context/prune ✓ excluded: repetition · tangents · unrelated chat · raw transcript
context/prune ✓ excluded: repetition · tangents · unrelated chat · raw transcript
SESSION-DERIVED IDEA BRIEFService Contract Intelligence MVP · proposed
ProblemContract review is slow, hard to verify, and easy to lose across handoffs.
Affected userContract-review team; the first exact role is still unresolved.
Desired outcomeUpload one synthetic service contract, see cited obligations and renewal dates, ask grounded questions, and record a human review.
Team outcomeReusable architecture, tooling recommendation, CI/CD starter, prompt guidance, and handover guide.
Constraints / non-goalsNo customer data, production integrations, or autonomous contract decisions.
Unresolved assumptionsFirst user role, source systems, legal review, security controls, and production operating model require evaluation.
source: pruned session contextraw transcript: excludedidea status: proposed
Scene 3 of 13 — Load idea planning
After building the pruned session-derived brief, /plan automatically loads develop-obsidian-idea.
Internal skill step · candidate-first discovery
soheragent[plan-concept] $run bounded idea discovery for the pruned session-derived briefinternal action
agent/loadLoad develop-obsidian-idea internallyready
agent/searchInspect bounded active ideas and initiative Inbox candidatescomplete
agent/matchNo canonical existing idea exactly matches this briefnew capture
boundaryDo not create specification, handoff, lease, or workerblocked
Scene 4 of 13 — Explain, then clarify
The agent explains the gap, its recommendation, and the effect before asking one question.
Clarification only · no approval requested
SoherAgent · action explanationclarification · user input needed
ObservedThe brief defines outcomes but does not identify the first user.
RecommendationStart with a service-contract reviewer who can verify every citation.
Effect of your answerNarrows workflow and acceptance checks only; grants no approval or execution.
Scene 5 of 13 — Apply the clarification
The terminal records the answer, then the agent derives a bounded recommendation.
Agent analysis · no approval or mutation
User answer recordedFirst user: Service-contract reviewerclarification only
soheragent[plan-concept] $derive scope from brief + recorded first-user answeragent action
Recommended MVP · synthetic intake, cited obligations and dates, grounded Q&A, human review
Recommended foundation · architecture, tooling, CI/CD, prompt, and handover starter
Explicit non-goal · customer data, production integration, or autonomous decisions
Authority remains off · implementation, handoff, commit, push, deploy
Scene 6 of 13 — Explain, then name
The agent explains why a durable name is needed before asking the user to choose.
Naming decision only · no approval
SoherAgent · action explanationnaming · user decision needed
ObservedThe brief and scope are clear, but the initiative has no durable identity.
RecommendationUse “Service Contract Intelligence MVP” to name both product intent and limited phase.
Effect of your choiceSets planning provenance only; it does not approve the idea or start work.
Scene 7 of 13 — Explain, then compare implementers
The agent records the name, checks the registry, and explains the choices before asking.
Preference only · no assignment or dispatch
User decision recordedWorking name: Service Contract Intelligence MVPnaming only
SoherAgent · registry explanationpreference · user input needed
ObservedThe reviewed shared registry lists Pi and Hermes for implementation; it has no Claude or Codex agent entry.
RecommendationPrefer Pi for this local workflow, then recheck runtime and workspace access at handoff.
Effect of your choiceStores a sticky preference only; assignment, authorization, and dispatch remain separate.
Scene 8 of 13 — Select Hermes as reviewer
The agent records Pi as implementer, recommends a separate reviewer, and focuses Hermes for the user to confirm.
Hermes focused · preference only · review cannot start
User preference recordedPreferred implementer: Pino assignment
SoherAgent · reviewer independencepreference · user input needed
ObservedPi is preferred for implementation; selecting the same agent for review would weaken independence.
RecommendationSelect Hermes as the separate reviewer, subject to availability and access checks before review.
Effect of your choiceStores Hermes as a sticky preference only; it does not authorize or begin review.
Scene 9 of 13 — Assemble the recommendation
The terminal records preferences, then the agent assembles one reviewable planning capsule.
Agent synthesis · idea still unapproved
User preference recordedPreferred reviewer: Hermes · separate from Pi implementerno review started
soheragent[plan-concept] $assemble reviewable capsule from brief + answers + preferencesagent action
AGENT RECOMMENDATION · INITIATIVE CAPSULE
NService Contract Intelligence MVPworking name
PCited contract insight with a human review recordMVP goal
TReusable architecture and engineering adoption guidefoundation
LSynthetic contracts only · no production integrationnon-goal
APi implementer · Hermes reviewerpreferences only
Scene 10 of 13 — Explain, then request authorization
The agent states its recommendation and exact authority boundary before the user decides.
Planning authorization · user owns disposition
SoherAgent · authorization briefinguser authorization required
ObservedA bounded capsule is ready; source-system, security, and legal details remain explicit discovery items.
RecommendationApprove only if this is enough for specification planning; otherwise re-evaluate with more questions.
Effect of your choiceApprove advances planning only. Re-evaluate loops. None of these options starts implementation.
Scene 11 of 13 — Record authority, then save
The terminal shows the submitted decision before parent-owned writes and verification.
Exact non-destructive writes · human approval shown
User authorization recordedDisposition: Approve idea · continue planning onlyexecution still off
parent[obsidian] $preview exact initiative + approved-idea checkpoint writesapproval gated
parent/previewExact capsule and idea changes shownreviewed
human/approveNon-destructive checkpoint writesapproved
parent/writeInitiative, idea disposition, and preferencessaved
parent/verifyCanonical paths, links, and approval provenancepassed
boundaryNo handoff, lease, worker, commit, push, or deployblocked
Scene 12 of 13 — Report the terminal result
The agent reports what changed, what happens next, and what remains forbidden.
Approved planning is not execution authority
soheragent[plan-concept] ›RESULT: idea checkpoint approved and verified. Continue internally to develop-obsidian-spec; ask again at the next user-owned decision.result
Changed · initiative capsule + idea approval + decision provenance
Next agent action · load develop-obsidian-spec internally
Still unauthorized · handoff, implementation, commit, push, deploy
Publication · false
you@pi:~$
Scene 13 of 13 — Open the detailed idea in Obsidian
The canonical idea opens at the top with metadata, nine sections, an outline, linked artifacts, and visible scroll continuation.
Detailed sanitized Obsidian recreation · planning only
