ssot-context-loop

SSOT Context Loop

Maintain a governed PMM knowledge system for one client or project. It separates approved strategy from living field intelligence, detects drift, records decisions, and never changes canonical strategy without human approval.

When to Use

Use this skill to create or maintain ssot/<client-slug>/, scan new field or market inputs, review drift, audit generated assets, or manage canon decisions. Load references/ssot-spec.md before acting. It is the detailed operational specification for the file schemas, commands, signal model, gates, and templates.

Relationship to Other Skills

• product-marketing-context establishes understanding and assembles proposals at .agents/product-marketing-context.md. It remains separate. If no SSOT exists, it may serve as context. If an SSOT exists for the active client or project, the SSOT is more authoritative for claims it covers.
• Never automatically overwrite .agents/product-marketing-context.md.
• positioning-strategy may develop strategy, but canon changes enter this SSOT only through its evidence and approval gates.
• messaging-framework should source differentiators from core and persona messaging from SSOT persona research. Do not create either standalone.
• icp-and-buyer-personas and customer-research can inform persona and living evidence files. competitor-profiling can inform competitive and market evidence.
• sales-enablement, marketing-copy, launch-strategy, market-entry-orchestration, and launch-readiness-check can read mapped SSOT files to keep assets aligned.
• events can adapt approved messages for event context and routes material narrative conflicts through this SSOT's evidence and human-approval gates.
• output-quality-check checks a skill's artifact against that skill's declared requirements. It does not audit canon consistency or change this SSOT.
• Use repo-level manifest.yaml in the user's project to map skills or agents to the SSOT files they read. Start with /manifest to create or update it.
• The SSOT is upstream context. Other skills do not call this skill to decide canon, and this skill does not depend on their outputs as authority by default.

Operating Rules

• Separate facts, which are recorded events, from opinions, which are strategy. Document opinions and ground them in facts.
• Never invent a claim. Cite the quote or data, speaker and role, account type, system, date, and outcome or cost when known.
• Every content line in every canonical SSOT file must end with exactly one evidence tag: [C], [A], or [G]. [C] is cited, [A] is assumed, and [G] is a gap. Apply decision-specific readiness: hold consequential unsupported decisions, continue independent work, and retain weak-evidence hypotheses with the evidence needed to resolve them. Approval remains separate.
• Core holds its existing strategy domains; scoped approved decisions/references cover topics without a core home, per the spec. Initial specialist proposals may use supplied evidence before canon exists. Existing canon remains authoritative.
• Give each claim one canonical home and reference dependencies. Update only explicitly approved affected sections, which may span files. Record uncertain placement in 09-iteration.md.
• Field signals inform the canon, but do not automatically change it. Every finding gets one proposed verdict and waits at the required human gate.
• Never write to core without Gate 2 approval. Never edit or delete comms/ input. Deprecate a section in place and name its replacement instead of deleting it.
• Follow the reference's file schema, staleness thresholds, taxonomy, and command outputs. state.md is written only by the reconciler, never by hand.
• Output without em dashes. Keep sentences to two lines at most. Lead bullets with outcomes and remove filler.

Command Routing

Route explicit commands to the matching workflow in references/ssot-spec.md:

• /setup <client> creates the baseline SSOT from supplied source documents.
• /scan <client> archives new inputs, classifies signals, and reports drift.
• /confirm <IDs> and /reject <ID> <reason> perform Gate 1 decisions.
• /propose prepares confirmed changes and impact analysis for Gate 2.
• /approve <IDs> and /revise <ID> <notes> handle Gate 2 decisions.
• /stale, /state, /brief, /manifest, and /audit-output produce their specified reports or project files.

For reviewed context proposals, read ../_shared/context-establishment.md and follow the spec's intake path. Preserve provenance and incoming IDs. Baselines wait at Gate 2; existing-context updates pass Gate 1 then Gate 2. Evidence readiness and human approval are separate; partial approval does not settle held decisions. For unstructured goal/material intake without a command, route to PMC to establish understanding. Resolve only missing scope. Never infer approval from silence.

Output Behavior

Write or propose files in the user's working project, never inside this skill or the PMM-Skills repository. Keep raw uploaded material in the project's immutable ssot/<client-slug>/comms/ directory. Respect the workflow's stop points.

After each workflow-authorized file creation or change, commit and push the generated SSOT output to the project's GitHub-backed repository. GitHub is the durable source of truth, not the chat. Preserve the required stop points: /scan stops at Gate 1, and proposed changes wait at Gate 2. Never modify core or commit a core change before explicit human approval at Gate 2.

For Gate 2 approval, return each ready-to-commit file in full, one code block per file with its path as the heading. Include the decision record for every core change, regenerated state.md, and a commit message. Commit and push the approved files after approval.

For /scan, return the drift report, three one-line priority findings, and "Needs your call" questions, then stop at Gate 1. For /brief, keep the team summary to ten lines or fewer and omit internal gate details.