influence-and-negotiation

Read references/memory.md for the full memory system. Then ask:

"Is this a continuation of an ongoing negotiation? If yes, do you have a memory document — an Artifact, Canvas, or file — from a previous session?"

If yes: ask the user to share the memory.md entrypoint. Spawn a sub-agent to read all referenced memory files per the load policy in references/memory.md and return their content to the main agent. Then read references/context-intake.md in incremental mode — collect new raw material only, run deep research only on new sources, pass the quality gate, resume from ## Next session plan in strategy.md.

If no: read references/context-intake.md and follow the three steps — collect raw material, run full deep research, and pass the quality gate — before proceeding to Phase 1.

Detect the mode AND the domain from the user's prompt. Domain cues:

• B2B sales: RFP, deal, ACV, procurement, ARR, champion
• Salary: raise, compensation, offer, counter-offer, equity, sign-on, band
• Social / annual collective bargaining: annual collective bargaining, union, CHRO, strike, works council
• Internal management: 1:1, performance plan, decision announcement, layoff, mediation
• Cross-cultural / diplomatic: international deal, M&A, joint venture, interpreter, protocol
• Recruitment: candidate, hire, offer close, back-channel, counter-offer

Domain shapes which axes matter and which references to load first; the workflow itself is the same.

For Preparation mode, run a live intake before anything else. Ask each question individually — don't dump them all at once. Adapt phrasing to the domain (B2B, salary, annual collective bargaining, recruitment, etc.).

Ask in this order, one at a time, and wait for the answer before continuing:

1. Stage — "Where are you in the process — early exploration, mid-negotiation, close to agreement, or post-verbal-yes?"
2. Stakes — "What's the size and scope here, and who's affected if this goes well or badly?"
3. Counterparty — "Who's at the table — names and roles? Is the decision-maker in the room, or is there someone off-stage?"
4. What's been said so far — "What are the last 2–3 things the counterparty said, as close to verbatim as you can get?" (Exact words carry signal that paraphrase loses — push for quotes.)
5. Authority limits — "What can you commit to without checking with anyone? Where's your escalation threshold?"
6. Walk-away — "At what point would you walk away from this entirely — what's your hard stop?"

Fuzzy answers reveal the mandate gap to fix first. If an answer is vague (e.g. "I don't know my walk-away"), surface that explicitly before proceeding — improvising on top of a fuzzy mandate produces the "More-More Syndrome" pathology where you over-ask at the moment of victory and lose the agreement in sight of the line.

Read references/prepare.md. Then ask to fill in any gaps from Phase 1 — don't assume what you don't know. Ask:

• Formal structure — "Who else is involved on their side? What's the decision-making chain — who approves, who can veto?"
• Informal influence — "Who do people defer to in the room even if they don't have the title? Is there someone off-stage who'll influence the outcome?"
• Motivation per stakeholder — "What does [name] personally get if this goes well? What do they lose if it doesn't?"
• Process gaps — "What formal steps still need to happen — legal review, board sign-off, infosec, exec sponsor alignment, HR validation?"
• Alternatives — "What's their fallback if this doesn't close? Have they mentioned any other options or comparisons?"

Layer formal org chart + informal influence map — domain-specific stakeholder cast in references/prepare.md.

If the user names a champion or advocate, ask: "What concrete actions have they taken between meetings — have they proactively coordinated internally, shared information you didn't ask for, or moved things forward without prompting?" The 3-question commitment test lives in references/prepare.md#champion-test. Skipping this validation is the highest-leverage error in complex negotiations.

Stakeholder deep research. Once stakeholders are named, run parallel sub-agents (one per person) to profile each across CRM, Slack, LinkedIn, and OSINT — see references/prepare.md#stakeholder-mapping--org-chart--influence-map for the full sub-agent protocol, source-tracking rules, and output format.

Read references/prepare.md. Then guide the user through the mandate axis by axis — don't hand them a template to fill in alone.

Start by asking: "What are the axes you're negotiating? List everything on the table — price, payment terms, timeline, scope, SLAs, equity, leave, title, etc."

Then, for each axis the user names, ask through the 5 Mandascan points:

• "What's your opening number / position for [axis]?" (Entry)
• "What would a great outcome look like for [axis]?" (Ideal)
• "What's your realistic internal target — what you'd genuinely commit to?" (Objective)
• "At what point would you need to pause and check with someone before agreeing on [axis]?" (Escalation/bascule)
• "What's your hard walk-away on [axis] — below this, no deal?" (Rupture)

Fuzzy Rupture = mandate gap. Derive it from BATNA: "If this fails, what's your next best option?" — that sets the floor.

After the mandate, POS the counterparty per axis — see references/prepare.md. Axes by domain and worked examples also in references/prepare.md.

BATNA sizes Rupture, then put it away — see references/prepare.md.

BATNA market research. Run 6 parallel sub-agents across CRM, Slack, and open sources (benchmarks, competitor pricing, regulatory constraints, alternative supply) to ground BATNA in data — see references/prepare.md#batna-zopa-and-the-operational-divergence for the full agent list and output format.

Read references/tactics.md NOW. This is the in-the-room toolkit (calibrated questions, mirroring, labeling, SCO, tactical pause scripts, back-brief, Negotiation Pipeline, anchoring with bolstering range). Do not draft scripts or pre-write moves without it — the specific phrasing matters.

Pre-write each artifact before the meeting; canonical phrasing in references/tactics.md:

• Opening anchor (bolstering range for salary asks; non-round numbers)
• Concession ladder (3–4 concessions, each paired with a counter-ask, one-for-one)
• 5–6 calibrated questions ("what" / "how", never "why")
• Accusation-audit labels that disarm objections before they form
• SCO statement — references/tactics.md#sco
• Tactical pause triggers + script — pre-decide signals and break script

Mutual Action Plan (where applicable). For mid-stage commercial deals, recruitment with multi-step approvals, or any negotiation with hidden gating steps, draft a MAP — see references/playbooks.md#map. It surfaces the legal / infosec / board / compliance-review / HR-validation steps that otherwise hide and creates joint ownership of the timeline. Stalls become diagnostic.

Team negotiation preparation. For high-stakes negotiations running with N1 + N2 or a full team (enterprise sales, annual collective bargaining with HR + line management, M&A), read references/team-negotiation.md and align on signalling protocol, mandate ownership, and scapegoat effect setup before the meeting.

When both MAP and team preparation apply, spawn two parallel sub-agents: one drafts the MAP using references/playbooks.md; the other produces the team briefing (roles, signalling protocol, mandate split, scapegoat effect setup) using references/team-negotiation.md. Both return full output to the main agent before Phase 5.

Number discipline. Specific anchor numbers and Mandascan figures belong in your private preparation notes — not in any counterparty-facing email, draft, or coaching artifact. Numbers leaked in writing become anchors for the other side or for your own commitment, and produce premature concessions. When coaching someone else, give them the strategic frame and the trade structure; let them say the number live on the call.

Pre-meeting competitive intelligence (B2B). For any commercial deal, run 6 parallel sub-agents across CRM, Slack, LinkedIn/Apollo, and open sources (current vendor signals, competitor positioning, buyer strategic signals, procurement history, analyst landscape, tech stack) — see references/prepare.md for the full agent list, source-tracking rules, and storage format.

Run a 3-minute mental simulation:

• Best objection — the one most likely to hit. Pre-write a label + reframe.
• Weakest objection — the one easiest to dismiss. Resist the temptation to spend cycles there.
• Surprise move — the gambit you didn't see coming (procurement escalation ladder, union ultimatum, manager pulling rank, candidate's current employer counter-offer). Pre-write a redirect.

The pre-mortem is the cheapest insurance against the "perte d'objectif" pathology — losing your mandate inside the room because you're improvising under stress.

Read the live-response references NOW, before drafting any response. Load references/tactics.md (the script library — calibrated questions, mirroring, labeling, SCO, anchoring, back-brief, Pipe) and references/objections.md, then navigate to the relevant objections section: refusal triage, four root objections, JOLT, procurement playbook, ghosting, non-negotiable, or face-saving exits. Do not improvise from the SKILL.md body alone — the specific scripts live in those files.

Triage the pushback type by reading references/objections.md#refusal-triage BEFORE drafting any reply — Emotional / Belief-based / Bad-faith / Identity-protective / Tactical each demand a different move; the reference has canonical signals and scripts.

For the four root commercial objections (price, timing, authority, no-need) and cross-domain equivalents, see references/objections.md#four-root.

No-decision diagnostic (JOLT). When the counterparty is engaged but not converging — saying yes to capability and no to commitment, or the deal stalls late without a substantive new objection — treat it as a no-decision case, not a loss to a competitor or a "needs more time" case.

The intervention is different: Judge / Offer / Limit / Take risk off — see references/playbooks.md#jolt. 40–60% of pipeline that doesn't close is no-decision; classical urgency tactics make it worse. The same pattern applies in promotion conversations (manager agrees in principle but never schedules HR sign-off) and in M&A (boards agree on strategic fit but defer signature indefinitely).

Manipulation taxonomy. When the counterparty's pushback fits a named manipulation pattern (bad faith, bluff, intimidation, punching-ball, faux pivot, feigned indifference, false cooperation, tactical silence, defeatism induction, closing manipulation), see references/manipulation.md for detection and counter-protocols that don't escalate.

Wrap-up before any agreement. Run a back-brief — see references/tactics.md. The counterparty reformulates each axis in their own words.

This is your defence against selective memory, closing manipulation, and genuine misunderstanding. At signature (or at the end of a salary conversation), run the Negotiation Pipeline closing checklist — see references/tactics.md.

Read references/debrief.md. Then guide the user through it — don't just describe the framework.

Step 1 — check emotions first. Ask: "Before we analyse what happened — how are you and the team feeling about it?" If the answer carries visible frustration or blame, run defusing before RetEx. Ask: "What happened that was hard? What are you still carrying from it?" Let it land, reflect it back, then move to facts.

Step 2 — RetEx, question by question. Ask to walk through each step:

1. "Walk me through the timeline of events — what happened, in order, as factually as you can?"
2. "Looking at those facts: what worked? Which tactics, moments, or scripts actually moved things?"
3. "What landed flat or created backlash? Where did you lose leverage you didn't need to lose?"
4. "If you ran this negotiation again from the same starting point, what would you change first?"
5. "What's transferable — what pattern would you teach to someone facing a similar situation?"

Step 3 — check for closing pathologies. Once the 5 RetEx answers are in hand, spawn a background agent: give it the complete RetEx narrative and instruct it to read references/debrief.md in full, then match the narrative exhaustively against all 5 pathology patterns (fear-of-failure, plan-b-preeminence, ego, "More-More Syndrome", target-fascination) and return a complete analysis — which patterns fired, the specific evidence from the narrative for each, and the recommended counter. The main agent continues the debrief conversation while this runs. When the background agent returns, surface its findings: if one or more patterns fired, name them directly — pattern recognition is 80% of the fix.

For drafted emails, scripts, or counter-proposals, invoke a humanizer skill (e.g. "humanize", "humanizer", "de-slop", "natural writing check", "AI detection cleanup") in the right language. AI-sounding prose triggers procurement scepticism, breaks champion trust, and undermines a difficult-conversation script that needs to land warm.

Preserve the calibrated questions and labels verbatim. They were tactically engineered (Phases 4 and 6); rewriting them for "naturalness" destroys the emotional logic. Tell the humanizer explicitly: keep questions and labels intact, scrub everything else.