fdeops 3.15.0 → 3.15.2

package/skills/fde/SKILL.md

@@ -1,23 +1,23 @@
 ---
 name: fde
-description: Keeps engagement memory for client work - sponsor, promise, what shipped, who accepted it. Use when the human names a client, customer or stakeholder. Use when they debrief a meeting or paste notes from one. Use when they ask what was agreed, or want dates and receipts. Use when they prep a client meeting or readout, when scope or trust shifts, or when they say @fde. Route the phase and run the local fde CLI (or npx --yes fdeops if it is not installed); never ask them to type commands. Not for ordinary code edits, unit tests, refactors or commits.
+description: Keeps the engagement record for client work. Use when they name a client or stakeholder. Use when they debrief a meeting or paste notes. Use when they ask what was agreed. Use when they run a POC, slice a feature on the client's codebase, prove it on their staging, or need evals before a model acts. Use when they prep a readout, when trust shifts, or they say @fde. Route and run the local fde CLI (or npx --yes fdeops). Never ask them to type commands. Not for ordinary code edits in an unbound repo.
 ---
 
 # @fde
 
 ## Purpose
 
-The **engagement record** for one client. One skill; six stages (land → close). You pick the method; they never pick a skill. Confirm, then write `.fde/`. The host agent writes the TypeScript; you log what they got. The artifact is the memory.
+The **engagement record** for one client, from first meeting to signed outcome. One skill; six stages (land → close). You route; they never pick a skill. Confirm, then write `.fde/`. The workspace still compiles and commits. `@fde` does not leave.
 
 ## When to use
 
 - They named a client, pasted notes, or asked what was agreed
 - The brief feels wrong, a sponsor went quiet, or Friday needs the ledger
-- Unbound — ask the name once, then **you** run `fde resume --init`
+- Unbound - ask the name once, then **you** run `fde resume --init`
 
 ## When NOT to use
 
-TypeScript errors, unit tests, refactors, git commits, generic debug: **host agent**. Agreed slice + code: implement in the host, then `fde log delivery`.
+A one-line typo or compile error in a file that will not ship. On a bound client: stay here for POC, slice, characterisation, eval, go-live, rollback, and acceptance.
 
 ## Use these first
 
@@ -25,14 +25,26 @@ TypeScript errors, unit tests, refactors, git commits, generic debug: **host age
 |---------|-----------------|---------|-----------|
 | **The brief is wrong** | "If this works, who in their company would have to agree that it worked?" | `fde resume` then discover | `references/discover.md` |
 | **They went quiet** | "Is this a process gap, or a trust problem?" | `fde log contact "…" --signal amber\|red\|green` | `references/rescue.md` |
-| **When did we agree?** | Don't argue from memory. Search the record. | `fde receipts <term>` | — |
-| **What did they get?** | A number nobody signed is claimed, not delivered. | `fde status` | `references/status.md` |
+| **When did we agree?** | Don't argue from memory. Search the record. | `fde receipts <term>` | - |
+| **What's the outcome?** | A number nobody signed is claimed, not delivered. | `fde status` | `references/status.md` |
 
 After a meeting: `fde debrief --smart` → confirm → `--apply`. Walk-in: `fde prep`. Friday: `fde status`.
 
+## Ground loop
+
+On someone else's site the work is not "write code, remember later." Every slice stays on `@fde`:
+
+1. **Name it** in `decisions.md` (plan) or kill it in a day (sketch).
+2. **Characterise their code** before you change it (incremental-build). Their tests, their runner.
+3. **Prove it where they live.** Staging they operate, a screen the signer in `success.md` can reject.
+4. **If a model judges:** `evals.md` Verdict SHIP before the slice is done (eval-pack).
+5. **Log delivery.** Outcome is promised → measured → accepted, not a green CI.
+
+A throwaway file can skip the loop. A client slice cannot.
+
 ## Human surface vs agent plumbing
 
-**FDE (human):** `@fde` + English, or `/brief` `/discover` `/plan` `/ship` `/got` `/close` `/debrief` `/prep` `/quiet` `/agreed` `/status`. Never a skill catalog.
+**FDE (human):** `@fde` + English, or `/brief` `/discover` `/plan` `/ship` `/outcome` `/close` `/debrief` `/prep` `/quiet` `/agreed` `/status`. Never a skill catalog.
 
 **You (agent):** run the CLI. **Never tell the FDE to type** `fde …`. If unbound, you run `fde resume --init` after one question. Never ask them to run the CLI.
 
@@ -42,7 +54,7 @@ Fallbacks: `node ~/.claude/fdeops/fde.js …`, then `npx --yes fdeops …`. Skil
 
 1. `fde resume` (bounded `context.md`). `--full` only if you need the whole log.
 2. **NO ENGAGEMENT:** ask "What should we call this client?" then **you** init. Pasted notes → debrief after bind.
-3. Playback 2–3 lines. `hygiene:` → offer `fde doctor`; **never auto-rewrite**.
+3. Playback 2-3 lines. `hygiene:` → offer `fde doctor`; **never auto-rewrite**.
 4. Route. Read **one** `references/*.md`. Confirm, then write.
 
 Writes need a bind (`FDEOPS_ENGAGEMENT` or registry). Never install fdeops on infrastructure they do not control.
@@ -54,10 +66,10 @@ Writes need a bind (`FDEOPS_ENGAGEMENT` or registry). Never install fdeops on in
 | debrief / pasted notes | `fde debrief --smart` → you rewrite prefixes → confirm → `--apply`. `--smart` is a gate, not a brain. `references/debrief.md` |
 | prep me for … | `fde prep "<label>"` |
 | when did we agree | `fde receipts <term>` |
-| sponsor update / what they got | `fde status` |
+| sponsor update / the outcome | `fde status` |
 | they went quiet | `fde log contact "…" --signal amber\|green\|red` |
 | fieldbook page | `fde dashboard` |
-| clean up the fieldbook | `fde doctor` — never auto-rewrite |
+| clean up the fieldbook | `fde doctor` - never auto-rewrite |
 | scrub a secret | `fde redact <term>` then `--apply` after confirm |
 | pull Granola/Slack/transcript | capability check → `fde ingest stage` → confirm → apply. Never auto-apply. `references/ingest.md` |
 | connect an MCP | `references/ingest-connect.md` |
@@ -69,35 +81,35 @@ Writes need a bind (`FDEOPS_ENGAGEMENT` or registry). Never install fdeops on in
 2. **Deliverable = memory.** The work *is* the `.fde/` file. The reference names which one.
 3. **Evidence.** Every claim has a source. Traceable beats plausible.
 4. **No invented facts.** People, quotes, meetings, numbers: they said it or the repo shows it. Else `unknown - ask: <question>`.
-5. **Session digest** (end of session and before a PR) — thinking, not the chat. Confirm, then write. Never a transcript dump.
+5. **Session digest** (end of session and before a PR) - thinking, not the chat. Confirm, then write. Never a transcript dump.
 
    | Digest beat | Lands in |
    |-------------|----------|
    | **TL;DR** | `context.md` |
-   | **Key decisions & why** | `decisions.md` — skip if none |
+   | **Key decisions & why** | `decisions.md` - skip if none |
    | **Pivot / aha** | `context.md` or `decisions.md` |
    | **Scope + verification** | `delivery.md` if code/PR; else skip |
    | **Gotchas** | `context.md` |
-   | **Next action** | existing `## Next action` — **replace**; never append a second heading |
+   | **Next action** | existing `## Next action` - **replace**; never append a second heading |
 
    Judgment ships in the fieldbook. Raw transcripts stay on the machine. The `session-stop` hook is a thin backstop; **you** write the digest.
 
 6. **One customer, one folder.**
 7. Never drop `## Signal history` or `## Retired` when rewriting those files.
 
-**Don't invent.** Don't tell them to run the CLI. Don't fill `success.md` / `terrain.md` with guesses. Don't ship on "probably fine" — intent vs diff, then pre-blast. Don't grill mid-flow. Don't sync transcripts into git.
+**Don't invent.** Don't tell them to run the CLI. Don't fill `success.md` / `terrain.md` with guesses. Don't ship on "probably fine" - intent vs diff, then pre-blast. Don't grill mid-flow. Don't sync transcripts into git.
 
 ## Data boundary
 
-CLI is local (`git` + files, no network). You see their code only when they point you at it. AI policy unknown → ask before loading code. `<private>` is redacted from CLI/dashboard/hooks — do not open raw private blocks with file tools.
+CLI is local (`git` + files, no network). You see their code only when they point you at it. AI policy unknown → ask before loading code. `<private>` is redacted from CLI/dashboard/hooks - do not open raw private blocks with file tools.
 
 ## Voice
 
-Direct. Their words. No "Certainly." Playback 2–4 lines, then act. One question only when a missing fact changes the next move.
+Direct. Their words. No "Certainly." Playback 2-4 lines, then act. One question only when a missing fact changes the next move.
 
 New embed: sprint / standard / programme changes depth, not which methods exist. Before first code: safe place to break things, plus AI-code policy. Before go-live: who needs to know, what's the rollback. Before a sponsor artifact: as-is or gut-check first.
 
-Muddy signal: name it ("discover or rescue — leaning X"). Never a phase-picker interview. Default: land if new, audit if takeover.
+Muddy signal: name it ("discover or rescue - leaning X"). Never a phase-picker interview. Default: land if new, audit if takeover.
 
 ## Routing - 6 stages
 
@@ -120,7 +132,7 @@ Read **one** reference and follow it. Do not improvise from memory.
 | Don't know the real problem, brief feels wrong, shadow processes | discover | `references/discover.md` |
 | The brief feels too neat, assumptions untested, "we just need…" | assumption-audit | `references/assumption-audit.md` |
 | Multiple use cases competing, "we want to do everything" | use-case-scoring | `references/use-case-scoring.md` |
-| Need to validate a direction, prototype, demo to de-risk | sketch | `references/sketch.md` |
+| Need to validate a direction, prototype, demo to de-risk, **POC**, spike, killer assumption | sketch | `references/sketch.md` |
 
 ### Plan
 
@@ -135,13 +147,13 @@ Read **one** reference and follow it. Do not improvise from memory.
 
 | You hear | Skill | Reference |
 |----------|-------|-----------|
-| Large feature, need visible progress every 2–3 days | incremental-build | `references/incremental-build.md` |
+| Large feature, need visible progress every 2-3 days, slice it, characterise their tests, POC follow-through | incremental-build | `references/incremental-build.md` |
 | What could go wrong, touching shared infrastructure, need to assess impact | blast-radius | `references/blast-radius.md` |
 | Production down, urgent - OR stakeholder gone quiet, trust slipping | rescue | `references/rescue.md` |
 | Ready to deploy, going live, pre-flight check | ship | `references/ship.md` |
 | Review this change, is it safe, does it match what we agreed | review | `references/review.md` |
 | Diff grew / scope creep in the PR / "did we only build what we said" / KEEP JUSTIFY SPLIT DROP | review (+ ship if going live) | `references/review.md` Stage 1 · `references/ship.md` Intent vs diff |
-| Wrap the session / share the thinking / catch teammates up / before I open the PR | (memory contract — session digest) | SKILL.md **On exit** — write TL;DR + decisions/why into `.fde/`; no transcript sync |
+| Wrap the session / share the thinking / catch teammates up / before I open the PR | (memory contract - session digest) | SKILL.md **On exit** - write TL;DR + decisions/why into `.fde/`; no transcript sync |
 | "We can always revert" - need to actually test the escape route | rollback-drill | `references/rollback-drill.md` |
 
 ### Prove
@@ -173,7 +185,7 @@ Read **one** reference and follow it. Do not improvise from memory.
 | Signal | Overlay |
 |--------|---------|
 | AI, ML, LLM, model, embeddings, RAG, agents, fine-tuning, inference, drift | `references/ai.md` |
-| Golden set, eval suite, eval pack, pass/fail before AI ship, HITL gate for model | `references/eval-pack.md` (+ `ai.md`) |
+| Golden set, eval suite, eval pack, pass/fail before AI ship, HITL gate for model, POC the model | `references/eval-pack.md` (+ `ai.md`) |
 | Deck, slides, report, governance framework, compliance pack, ADR, PDF | `references/artifacts.md` |
 | Patient data, PHI, HIPAA, EHR, clinical | `references/healthcare.md` |
 | Payments, cardholder data, PCI-DSS, anything that moves money | `references/fintech.md` |
@@ -184,8 +196,9 @@ Ready to build with no `terrain.md` / plan: discover or plan first. Takeover wit
 ## Principles
 
 - Never ask the FDE to pick a phase. That's your job.
-- Read `context.md` before speaking. One sharp question — never a barrage.
-- Never invent people, meetings, or numbers — `unknown - ask:` beats a polished lie.
+- Ground loop on a bound client: name → characterise → prove where they live → log. Do not hand the slice to a generic coding pack.
+- Read `context.md` before speaking. One sharp question - never a barrage.
+- Never invent people, meetings, or numbers - `unknown - ask:` beats a polished lie.
 - Every phase ends with its artifact written. No artifact, no "done."
 - Evidence on every claim. The FDE will be challenged on these files.
 - Overlays activate on signal, not on request.

package/skills/fde/references/ai.md

@@ -29,7 +29,7 @@ Never start with the most powerful model. Start with the cheapest that meets the
 4. **Does it need fine-tuning?** Only when: you have 500+ high-quality examples, the base model fails consistently on your domain, and the cost of inference at scale justifies the training cost.
 
 **Evaluation method (before choosing):**
-- Build a test set: 50–100 representative inputs with expected outputs.
+- Build a test set: 50-100 representative inputs with expected outputs.
 - Run every candidate model against the test set.
 - Score: accuracy, latency, cost per call, failure modes.
 - The cheapest model that scores above the quality threshold wins.
@@ -38,14 +38,14 @@ Write model selection rationale to `decisions.md`. Include: models tested, test
 
 ## Engagement eval pack (before AI ships)
 
-When any slice touches a model, embeddings, RAG, or an agent: create or update `.fde/evals.md` **before** ship. Full method: `references/eval-pack.md`. This is the engagement-local test set — not unit tests.
+When any slice touches a model, embeddings, RAG, or an agent: create or update `.fde/evals.md` **before** ship. Full method: `references/eval-pack.md`. This is the engagement-local test set - not unit tests.
 
 **Minimum pack (do not grow until the minimum exists):**
-1. **Component + quality bar** — one sentence each; kill switch / fallback named.
-2. **Golden cases** — 5–20 representative inputs with expected outputs and a pass rule. Prefer real production-shaped data (sanitized).
-3. **Failure modes** — at least the silent ones: hallucination/ungrounded, retrieval miss (if RAG), drift, cost runaway.
-4. **Pass/fail** — dated run; Verdict **SHIP** or **NO-SHIP**; critical fails must be 0.
-5. **HITL gate** — which decisions need human review before action (align with `trust-profile.md`). Empty when policy requires review → NO-SHIP.
+1. **Component + quality bar** - one sentence each; kill switch / fallback named.
+2. **Golden cases** - 5-20 representative inputs with expected outputs and a pass rule. Prefer real production-shaped data (sanitized).
+3. **Failure modes** - at least the silent ones: hallucination/ungrounded, retrieval miss (if RAG), drift, cost runaway.
+4. **Pass/fail** - dated run; Verdict **SHIP** or **NO-SHIP**; critical fails must be 0.
+5. **HITL gate** - which decisions need human review before action (align with `trust-profile.md`). Empty when policy requires review → NO-SHIP.
 
 **When to write:** plan seeds the pack; sketch/build grows goldens; ship requires Verdict SHIP and a receipt in `delivery.md` → `## Ship receipts`. Non-AI work skips this file entirely.
 
@@ -59,7 +59,7 @@ When the AI needs to answer questions about the client's data:
 3. **Generate** - chunks + query → LLM → answer with citations
 
 **Common failure modes:**
-- **Chunk size wrong.** Too small = lost context. Too large = noise drowns signal. Start at 500–1000 tokens with 100-token overlap.
+- **Chunk size wrong.** Too small = lost context. Too large = noise drowns signal. Start at 500-1000 tokens with 100-token overlap.
 - **No citation/grounding.** If the model can't point to where it found the answer, you can't verify it. Always require source attribution.
 - **Stale index.** Documents update, embeddings don't. Define the refresh cadence. Real-time for critical data, daily for reference docs.
 - **Retrieval miss.** The right document exists but wasn't retrieved. Test with known-answer queries where the answer IS in the corpus - if retrieval misses these, the embedding model or chunking strategy needs work.

package/skills/fde/references/assumption-audit.md

@@ -76,11 +76,11 @@ Tell the FDE: how many assumptions extracted, how many critical, which ones were
 
 Acme's brief reads cleanly, which is the signal.
 
-Extracted assumptions include one nobody said aloud: *finance would act on an alert*. The whole plan rests on it, and the evidence behind it is a sentence in a kickoff. Blast radius CRITICAL — if false, alerting changes nothing and the engagement delivers a page nobody answers.
+Extracted assumptions include one nobody said aloud: *finance would act on an alert*. The whole plan rests on it, and the evidence behind it is a sentence in a kickoff. Blast radius CRITICAL - if false, alerting changes nothing and the engagement delivers a page nobody answers.
 
 Validation is a test, not a discussion, and it is cheap: send one real failure notification to the finance channel and watch what happens. It goes first because highest blast radius × cheapest test is the killer test.
 
-Result: acked in 40 minutes, by Marco, not finance. Assumption DISPROVED, and the plan changes before six weeks are spent on it — the alert needs a rota with an owner, which is a different piece of work than the one that was funded. `assumptions.md` records the status, the evidence, and the date; the finding is presented to the FDE as a fact base, not as "the brief was wrong".
+Result: acked in 40 minutes, by Marco, not finance. Assumption DISPROVED, and the plan changes before six weeks are spent on it - the alert needs a rota with an owner, which is a different piece of work than the one that was funded. `assumptions.md` records the status, the evidence, and the date; the finding is presented to the FDE as a fact base, not as "the brief was wrong".
 
 ## Principles
 

package/skills/fde/references/blast-radius.md

@@ -26,7 +26,7 @@ CONTAINED  → Only the module you're changing is affected
              Rollback: revert the PR
              Example: changing a utility function with no external callers
 
-ADJACENT   → 2–5 callers or one downstream system affected
+ADJACENT   → 2-5 callers or one downstream system affected
              Rollback: revert the PR + verify downstream
              Example: changing an API response format used by the frontend
 

package/skills/fde/references/business-case.md

@@ -30,7 +30,7 @@ Keep the drivers explicit. "We estimate $200K savings" means nothing. "3 people
 
 **3. Sensitivity check - name the two drivers that swing the result:**
 
-Every business case has 1–2 variables where a small change flips the outcome. Name them explicitly:
+Every business case has 1-2 variables where a small change flips the outcome. Name them explicitly:
 
 > "This case holds if the team actually reclaims 6+ hours/week per person. If it's only 3 hours, the payback extends from 5 months to 14 months. The validation: measure time-spent before and after pilot with 2 team members."
 
@@ -75,9 +75,9 @@ Acme phase 2 needs funding. The case starts with the cost of doing nothing, not
 
 Anchor: two silent failures since March, each one day of finance reconciliation by hand plus a late close (`reality.md`, Marco's sheet). That is the number the sponsor already believes because her own team reported it.
 
-Driver model the sponsor can trace: incidents/quarter × hours of manual reconciliation × loaded cost, plus the tail risk of a late regulatory close — stated separately, because mixing a certain small number with an uncertain large one is how a case loses credibility.
+Driver model the sponsor can trace: incidents/quarter × hours of manual reconciliation × loaded cost, plus the tail risk of a late regulatory close - stated separately, because mixing a certain small number with an uncertain large one is how a case loses credibility.
 
-Sensitivity names the two drivers that swing it: incident frequency (2/quarter → 1/quarter and the case halves) and whether the manual re-run continues in parallel (if Marco keeps re-running every morning, the saving is theoretical). The second one is the honest weakness, so it is in the case rather than waiting to be found in the room — with the condition that makes it hold: the morning re-run stops after two clean cycles, agreed with Marco.
+Sensitivity names the two drivers that swing it: incident frequency (2/quarter → 1/quarter and the case halves) and whether the manual re-run continues in parallel (if Marco keeps re-running every morning, the saving is theoretical). The second one is the honest weakness, so it is in the case rather than waiting to be found in the room - with the condition that makes it hold: the morning re-run stops after two clean cycles, agreed with Marco.
 
 ## Principles
 

package/skills/fde/references/close.md

@@ -18,7 +18,7 @@ The engagement doesn't end at ship. It ends when the customer can maintain what
 - AI components: did they behave in production? What failure modes did the prototype hide? Is the team equipped to maintain them?
 
 **1b. Value + receipts close gate (refuse green close if any fail):**
-- Primary value bucket in `success.md` matches what the sponsor funded; at least one ledger row has **Measured** (not forever-`pending`) with evidence **and a named customer-side owner in Accepted by** for that bucket — or the retrospective explicitly records “not measured; sponsor accepted pending.” A measured-but-unaccepted number closes as `claimed`; say so in the retrospective rather than closing green on arithmetic nobody signed.
+- Primary value bucket in `success.md` matches what the sponsor funded; at least one ledger row has **Measured** (not forever-`pending`) with evidence **and a named customer-side owner in Accepted by** for that bucket - or the retrospective explicitly records “not measured; sponsor accepted pending.” A measured-but-unaccepted number closes as `claimed`; say so in the retrospective rather than closing green on arithmetic nobody signed.
 - Audit receipt exists for the final shipped path (exceptions/operating map walked; cite file).
 - Eval receipt: **n/a if no AI**, else final golden/eval result + HITL owner recorded; kill switch / fallback named in `handoff.md`.
 - One line in the retrospective: which bucket moved, by how much, vs baseline.
@@ -39,17 +39,17 @@ The engagement doesn't end at ship. It ends when the customer can maintain what
 
 ## Checkpoint
 
-Direct assessment to the FDE: did the engagement achieve `success.md` · 2–3 lessons that matter · is the pattern worth encoding · is the handoff complete or where are the gaps. Also: value bucket + audit receipt green; eval **n/a or green**. Pending Measured without sponsor acceptance = gap, not green close. Honest - a gap named now is cheaper than a callback in six weeks.
+Direct assessment to the FDE: did the engagement achieve `success.md` · 2-3 lessons that matter · is the pattern worth encoding · is the handoff complete or where are the gaps. Also: value bucket + audit receipt green; eval **n/a or green**. Pending Measured without sponsor acceptance = gap, not green close. Honest - a gap named now is cheaper than a callback in six weeks.
 
 ## Worked example
 
 Acme, twelve weeks in, the FDE is rolling off.
 
-Retrospective against the receipts: `brief.md` asked for monitoring, `reality.md` proved it was ownership — and the delta is the most useful paragraph in the file, because it is exactly the argument the next engagement will need.
+Retrospective against the receipts: `brief.md` asked for monitoring, `reality.md` proved it was ownership - and the delta is the most useful paragraph in the file, because it is exactly the argument the next engagement will need.
 
-The close gate bites in a useful way. The ledger shows detection at 12 minutes measured across two real incidents, but **Accepted by** is empty — Marco confirmed it in Slack, Denise (finance) never did, and Denise is whose escalation started the engagement. So it closes as `claimed` with a one-line retrospective note and a named next step, rather than a green close on a number nobody with budget agreed to.
+The close gate bites in a useful way. The ledger shows detection at 12 minutes measured across two real incidents, but **Accepted by** is empty - Marco confirmed it in Slack, Denise (finance) never did, and Denise is whose escalation started the engagement. So it closes as `claimed` with a one-line retrospective note and a named next step, rather than a green close on a number nobody with budget agreed to.
 
-`handoff.md` is written for the person woken at 2am: the three things that break, what the page means, how to re-run manually the way Marco does, and who holds the tribal knowledge (Raj, who built the original job — credited, because he protects it now). `patterns.md` gets *"unowned job" presents as "unmonitored job"* — it has now happened twice.
+`handoff.md` is written for the person woken at 2am: the three things that break, what the page means, how to re-run manually the way Marco does, and who holds the tribal knowledge (Raj, who built the original job - credited, because he protects it now). `patterns.md` gets *"unowned job" presents as "unmonitored job"* - it has now happened twice.
 
 ## Principles
 

package/skills/fde/references/debrief.md

@@ -12,7 +12,7 @@
 
 - The `fde` CLI is **local, deterministic, no AI**. `--smart` is a **gate + writer**, not a brain.
 - It keeps lines that already have `decision:` / `risk:` / `delivery:` / `contact:` / `next:` prefixes, plus a thin keyword pass (e.g. "we agreed", person+verb lines, "open question").
-- Real messy notes without prefixes often route **0 useful lines** — everything else lands as a context dump. That is expected. **You are the router:** rewrite `.debrief-propose` with type prefixes, then `--apply`.
+- Real messy notes without prefixes often route **0 useful lines** - everything else lands as a context dump. That is expected. **You are the router:** rewrite `.debrief-propose` with type prefixes, then `--apply`.
 - `.debrief-propose` is raw lines only (no routing annotations). "Edit if mis-routed" means **rewrite the line with the right prefix**, not leave a comment in the file.
 
 ## Method (you do this work)
@@ -22,7 +22,7 @@
 1. Save the FDE's notes to a temp `.md` file in the workspace (or pipe stdin).
 2. Run `fde debrief --smart <notes.md>` (or `npx fdeops debrief --smart …`).
 3. Open `.debrief-propose`. If lines lack type prefixes, **rewrite them** before showing the FDE, e.g.:
-   - `decision: agreed chargebacks stay phase 2 — Priya`
+   - `decision: agreed chargebacks stay phase 2 - Priya`
    - `risk: legal may reopen scope if we slip the SOW date`
    - `contact: Priya pushed hard on Friday deck [signal:amber]`
    - `next: send one-pager before Thursday 9am`

package/skills/fde/references/discover.md

@@ -22,16 +22,16 @@ State your read, let the FDE correct, then discover.
 
 Use when the "problem" is unfalsifiable, success is undefined, or you cannot name the decision discovery informs. Skip when `reality.md` / `terrain.md` already pin a testable claim and the FDE is ready to dig.
 
-Same format as land — one Q + GUESS, no checklist:
+Same format as land - one Q + GUESS, no checklist:
 
 ```
 READ: <the real problem you think exists, in one sentence>
-CONFIDENCE: ~NN% — missing: <what would falsify or confirm it>
+CONFIDENCE: ~NN% - missing: <what would falsify or confirm it>
 Q: <one question that changes where you dig>
 GUESS: <your answer, so they can correct it>
 ```
 
-Stop when you can write the decision sentence under **Frame the decision first**. If a name, quote, or metric is still missing, write `unknown - ask:` — never invent ops folklore to make the map look complete.
+Stop when you can write the decision sentence under **Frame the decision first**. If a name, quote, or metric is still missing, write `unknown - ask:` - never invent ops folklore to make the map look complete.
 
 ## Frame the decision first
 
@@ -92,7 +92,7 @@ The real spec is what people **do** when the system fails - not what the slide d
 - **The hesitation.** When someone says "well, there's also this other thing we do…" - stop them, ask them to finish. The main story is what they're comfortable explaining; the hesitation is the real problem.
 - **"Which part of the codebase do you least want to touch?"** The answer is unanimous and it's the load-bearing wall. Check it against your churn scan - when the human answer and the churn data agree, that's your first map landmark.
 - **Shadow AI.** Someone pasting data into ChatGPT to cope = a real unmet need + an uncontrolled data risk. Note both.
-- **Exception-led operating map.** For each real break (not the slide-deck process): what fails, who notices first, what they do today, and which artifact is trusted in that moment. Prefer exceptions over happy-path swimlanes — the workaround is the operating system. Write rows under `terrain.md` → `## Operating map (exception-led)`. If the section is missing on an older engagement, add it; never regenerate the rest of terrain. When AI is in play, also fill `## Intelligence placement` (deterministic vs LLM judgement vs human approve). **`fde doctor` requires at least one filled exception row before plan/build/ship/close** — empty map after discover is a hygiene fail, not optional polish.
+- **Exception-led operating map.** For each real break (not the slide-deck process): what fails, who notices first, what they do today, and which artifact is trusted in that moment. Prefer exceptions over happy-path swimlanes - the workaround is the operating system. Write rows under `terrain.md` → `## Operating map (exception-led)`. If the section is missing on an older engagement, add it; never regenerate the rest of terrain. When AI is in play, also fill `## Intelligence placement` (deterministic vs LLM judgement vs human approve). **`fde doctor` requires at least one filled exception row before plan/build/ship/close** - empty map after discover is a hygiene fail, not optional polish.
 
 ## Method - part 3: workshop facilitation
 
@@ -137,7 +137,7 @@ A use case that depends on a "Blocker" source doesn't get scored - it gets a dat
 
 Score every candidate use case before anything gets prototyped:
 
-| Dimension | Question | 1–5 |
+| Dimension | Question | 1-5 |
 |---|---|---|
 | Business value | What does it cost them unsolved? | |
 | Complexity | How hard to build safely? (5 = hardest) | |
@@ -196,9 +196,9 @@ Stop. Don't form a fourth hypothesis. Three disproven reads means the brief is a
 
 Acme's brief blamed missing monitoring. Discovery goes to the workaround first.
 
-`git log` shows the reconciliation module at 47 commits/90d with no tests, all from one author who left in February. Marco (ops lead) turns out to keep a spreadsheet: every morning he re-runs the job manually and eyeballs the totals — a habit nobody mentioned because to him it is just the job. That spreadsheet is the system of record when the job fails, which is the actual finding.
+`git log` shows the reconciliation module at 47 commits/90d with no tests, all from one author who left in February. Marco (ops lead) turns out to keep a spreadsheet: every morning he re-runs the job manually and eyeballs the totals - a habit nobody mentioned because to him it is just the job. That spreadsheet is the system of record when the job fails, which is the actual finding.
 
-`reality.md`: **Confirmed:** the job has no owner, and the manual re-run masks failures for a day (evidence: Marco's sheet, Day 5; two silent failures since March, finance escalation Mar 14). **Stated brief was wrong because:** alerting existed last year and was disabled — adding it again without an owner reproduces the same outcome. `terrain.md` gets the hotspot row and an operating-map row: `job fails silently → Marco notices next morning → re-runs by hand → spreadsheet is truth → LOAD-BEARING (Marco, Day 5)`.
+`reality.md`: **Confirmed:** the job has no owner, and the manual re-run masks failures for a day (evidence: Marco's sheet, Day 5; two silent failures since March, finance escalation Mar 14). **Stated brief was wrong because:** alerting existed last year and was disabled - adding it again without an owner reproduces the same outcome. `terrain.md` gets the hotspot row and an operating-map row: `job fails silently → Marco notices next morning → re-runs by hand → spreadsheet is truth → LOAD-BEARING (Marco, Day 5)`.
 
 Checkpoint to the FDE names the sponsor decision this creates: fund ownership, or fund alerting and accept the same failure in six months.
 

package/skills/fde/references/eval-pack.md

@@ -1,6 +1,6 @@
 # eval-pack - prove the system before it acts
 
-**Enter when:** the work touches AI/LLM/agents/RAG, or ship/close is blocked because there is no evidence the non-deterministic path is safe. Activate alongside `ai.md`, `sketch`, `build`, or `ship` — not instead of them.
+**Enter when:** the work touches AI/LLM/agents/RAG, or they need to POC a model, or ship/close is blocked because there is no evidence the non-deterministic path is safe. Activate alongside `ai.md`, `sketch`, `incremental-build`, or `ship` - not instead of them.
 
 **Read first:** `trust-profile.md` (AI policy + HITL), `terrain.md` (operating map), `delivery.md`. Create or extend `evals.md`.
 
@@ -14,19 +14,19 @@ Intelligence without evidence is token-maxing with a nicer name. An FDE earns tr
 
 **1. Scope the judgement surface.** One sentence: which step uses model judgement, and what must never be autonomous.
 
-**2. Build a golden set (minimum 5–20 for a slice; prefer 50–100 before broad scale).** For each case:
-- input (sanitized — no `<private>` raw values)
+**2. Build a golden set (minimum 5-20 for a slice; prefer 50-100 before broad scale).** For each case:
+- input (sanitized - no `<private>` raw values)
 - expected outcome or expert-approved acceptance note
 - pass rule (exact / contains / short rubric)
 - source: real historical example / expert label / staged fixture
 
 **3. Score pass/fail, not vibes.** Run the suite. Record count pass / fail. Failures get a failure-mode tag (missing data, wrong record, format drift, hallucination, retrieval miss, unsafe action, other).
 
-**4. Human-in-the-loop gate.** Name which outcomes require human approve before side effects. If none, write why that is allowed under `trust-profile.md` AI policy — do not invent permission.
+**4. Human-in-the-loop gate.** Name which outcomes require human approve before side effects. If none, write why that is allowed under `trust-profile.md` AI policy - do not invent permission.
 
 **5. Ship rule.** Until `evals.md` shows Verdict **SHIP** with a dated run (critical fails = 0) and HITL filled when policy requires it, AI-touching ship stays **fix-first**. Log a one-line eval receipt in `delivery.md` → `## Ship receipts`.
 
-## Artifact — `evals.md`
+## Artifact - `evals.md`
 
 Create on first AI-touching slice (not at `resume --init`). Use the stub in `templates/.fde/evals.md`. Every claim needs a source. Missing evidence → leave the cell `unknown - ask:`, never invent scores.
 
@@ -38,6 +38,6 @@ Present to the FDE: suite size, pass rate, top failure mode, HITL gate, Verdict
 
 - No golden set, no AI ship.
 - Pass/fail beats “looks good.”
-- Failure modes are the product — the happy path is table stakes.
+- Failure modes are the product - the happy path is table stakes.
 - HITL is a gate, not a slide.
 - Non-AI work does not need this file.

package/skills/fde/references/handoff-engineering.md

@@ -58,13 +58,13 @@ Rollback: <exact command and expected time>
 
 | Session | Focus | Attendees | Duration | Output |
 |---------|-------|-----------|----------|--------|
-| **Architecture walkthrough** | Why, not what. The decisions, the trade-offs, the things that almost went wrong. | Full team | 60–90 min | Recording + Q&A log |
+| **Architecture walkthrough** | Why, not what. The decisions, the trade-offs, the things that almost went wrong. | Full team | 60-90 min | Recording + Q&A log |
 | **Operational drill** | Deploy, rollback, incident response. They do it, you watch. | On-call team | 60 min | Drill report with confidence level |
 | **Edge-case handover** | The things that aren't in any document. The workarounds, the fragile spots, the "ask Sarah because she's the only one who knows." | Team lead + 1 | 30 min | Additions to `handoff.md` |
 
 **4. The confidence check.** After the knowledge transfer, score the team's readiness:
 
-| Area | Confidence (1–5) | Evidence |
+| Area | Confidence (1-5) | Evidence |
 |------|------------------|----------|
 | Daily operations | | Can they deploy and rollback without help? |
 | Incident response | | Did they complete the drill within acceptable time? |

package/skills/fde/references/incremental-build.md

@@ -1,6 +1,6 @@
 # incremental-build - thin slices on someone else's codebase
 
-**Enter when:** the build task is larger than a single PR, multiple files or systems are involved, or the FDE needs to show visible progress to a stakeholder every 2–3 days.
+**Enter when:** the build task is larger than a single PR, multiple files or systems are involved, or the FDE needs to show visible progress to a stakeholder every 2-3 days.
 
 **Read first:** `decisions.md` (the plan), `terrain.md` (the danger zones), `context.md`. This skill works *inside* the build phase - it's the execution discipline that makes large features safe on codebases you don't own.
 
@@ -37,27 +37,36 @@ Each vertical slice delivers working functionality the customer can see. Each sl
 
 ```
 Read existing code in the area (search before creating)
-  → Write characterisation tests for what's there (if legacy)
+  → Characterise what is already there (their tests, their runner)
     → Implement the minimal working path
-      → Verify with evidence (tests + typecheck + smallest proving run)
+      → On-site proof (below)
         → Cleanup pass (dedupe, simplify - behaviour unchanged)
           → Self-review against acceptance criteria
-            → Commit with descriptive message
+            → Commit with a message the client's team can read
               → Update decisions.md + delivery.md
 ```
 
+**On-site proof.** A green check on your laptop is not delivery. Before the slice is done:
+
+- Run **their** test command, typecheck, or smallest proving path. Write the command and the result in `delivery.md`.
+- If the signer in `success.md` cannot reject this slice on a screen they already use, it is not proven.
+- Staging they operate beats a local demo. If you have no staging: `unknown - ask:` who owns an environment, then stop pretending it shipped.
+- Model in the path: `eval-pack` until `evals.md` says SHIP. Do not skip because "it looked right in chat."
+
+Do not prove it with a textbook ritual. The proof is whatever this client already believes, plus one new receipt they can replay.
+
 **4. Size discipline.** Each slice targets:
 
 | Metric | Target | Why |
 |--------|--------|-----|
-| Lines changed | 100–300 | Reviewable in one sitting |
-| Time to implement | 30–90 minutes | Testable before context decays |
-| Files touched | 1–5 | Blast radius stays containable |
+| Lines changed | 100-300 | Reviewable in one sitting |
+| Time to implement | 30-90 minutes | Testable before context decays |
+| Files touched | 1-5 | Blast radius stays containable |
 | Tests added | ≥1 per new behaviour | Proves the slice works; guards against regression |
 
 A slice larger than 300 lines → split before implementing. "It's all connected" means the design needs work, not the slice limit.
 
-**5. Stakeholder visibility rhythm.** Every 2–3 slices, something the customer can see:
+**5. Stakeholder visibility rhythm.** Every 2-3 slices, something the customer can see:
 
 - A working endpoint they can hit
 - A UI change they can click
@@ -80,12 +89,12 @@ Technical progress invisible to stakeholders is trust decay. `delivery.md` gets
 
 ## Checkpoint
 
-After each slice: tests pass (state the command and result), acceptance criteria met, blast radius as declared. After every 2–3 slices: stakeholder visibility confirmed - what did they see, and what's their signal?
+After each slice: tests pass (state the command and result), acceptance criteria met, blast radius as declared. After every 2-3 slices: stakeholder visibility confirmed - what did they see, and what's their signal?
 
 ## Principles
 
 - Vertical slices, always. Horizontal layers are untestable until assembled.
-- 100–300 lines per slice. Larger means split first.
+- 100-300 lines per slice. Larger means split first.
 - Every slice is independently revertible. If it isn't, the design is coupled.
-- Visible progress every 2–3 slices. Technical progress alone is trust decay.
+- Visible progress every 2-3 slices. Technical progress alone is trust decay.
 - The ugly code outside your slice stays ugly. That's discipline, not laziness.

package/skills/fde/references/ingest-connect.md

@@ -15,15 +15,15 @@
 
 ## Method
 
-1. **Ask one question** — which source? (`file` / `granola` / `slack` / `notion` / other). If "other", ask for the MCP they intend to use. If they just want paste → send them to debrief and stop.
-2. **Capability check (current session)** — list MCP tools you can actually call:
+1. **Ask one question** - which source? (`file` / `granola` / `slack` / `notion` / other). If "other", ask for the MCP they intend to use. If they just want paste → send them to debrief and stop.
+2. **Capability check (current session)** - list MCP tools you can actually call:
    - Sink: `fde ingest` CLI (preferred) and/or `ingest_stage`
    - Source: anything that can **fetch** that system's content (not post)
    - Say clearly: *available now* vs *needs config*.
-3. **Emit config for the source only** — open `mcp/recipes/<source>.md`. Fill placeholders from *that product's* docs. Tell them to paste secrets into host env — never into `.fde/`.
-4. **Tell them where to paste** — Cursor MCP settings / `mcp.json`. Claude Code: their MCP config. Save → reload MCP / restart session.
-5. **Verify** — after reload: re-run capability check. If source tools appear, offer a **test pull** staged to `.inbox/` only. Stop before apply unless they ask to propose.
-6. **Handoff phrase** — e.g. `@fde pull today's Acme Granola into the fieldbook`.
+3. **Emit config for the source only** - open `mcp/recipes/<source>.md`. Fill placeholders from *that product's* docs. Tell them to paste secrets into host env - never into `.fde/`.
+4. **Tell them where to paste** - Cursor MCP settings / `mcp.json`. Claude Code: their MCP config. Save → reload MCP / restart session.
+5. **Verify** - after reload: re-run capability check. If source tools appear, offer a **test pull** staged to `.inbox/` only. Stop before apply unless they ask to propose.
+6. **Handoff phrase** - e.g. `@fde pull today's Acme Granola into the fieldbook`.
 
 ## If they only want paste / files
 

package/skills/fde/references/ingest.md

@@ -1,6 +1,6 @@
 # ingest - pull large artifacts into the fieldbook loop
 
-**Enter when:** the FDE wants to catch the engagement up from external sources — "make sure Acme is up to date," "pull what's relevant," "grab today's Granola and Denise's last email." Raw transcripts and long emails that are too big to paste usefully.
+**Enter when:** the FDE wants to catch the engagement up from external sources - "make sure Acme is up to date," "pull what's relevant," "grab today's Granola and Denise's last email." Raw transcripts and long emails that are too big to paste usefully.
 
 **Connect / capability (different entry):** "connect a new MCP", "connect Granola/Slack/Notion", "what can you pull?" → `references/ingest-connect.md` first. Recipes: `mcp/recipes/` (file, granola, slack, notion).
 
@@ -11,7 +11,7 @@
 ## Honest contract (read once)
 
 - FDEOps owns the **sink only**: stage raw pulls → propose → confirm → apply. Nothing writes `.fde/` unreviewed.
-- **Source MCPs are the FDE's.** Granola, Slack, Notion, Gmail, custom — whatever they configured in Cursor/Claude. fdeops does not bundle OAuth, connectors, or ambient sync, and **does not push** to those tools.
+- **Source MCPs are the FDE's.** Granola, Slack, Notion, Gmail, custom - whatever they configured in Cursor/Claude. fdeops does not bundle OAuth, connectors, or ambient sync, and **does not push** to those tools.
 - Prefer **`fde ingest` in this bound workspace.** Optional `fdeops-ingest` MCP: pass `engagement` (path to `.fde/` from `fde resume --bind`) because MCP cwd often is not the client workspace.
 - The core `fde` CLI stays local (git + file reads). Source credentials live with that MCP; fdeops never stores them.
 - After apply, raw stays in `.inbox/`; the system of record (`.fde/`) stays thin dated facts.
@@ -20,20 +20,20 @@
 
 List what you can actually call **this session**:
 
-1. **Sink** — `ingest_stage` / `fde ingest` available?
-2. **Sources** — which fetch tools exist (Granola-shaped, Slack, Notion, Drive, file-only)?
+1. **Sink** - `ingest_stage` / `fde ingest` available?
+2. **Sources** - which fetch tools exist (Granola-shaped, Slack, Notion, Drive, file-only)?
 3. Tell the FDE in one line: *I can pull from X; Y is not connected.* If they asked to pull Y and it is missing → switch to `ingest-connect.md`. Never pretend a source exists.
 
 ## Ground loop (you do this work)
 
-1. **Bind** the engagement (`fde resume` / registry). If multiple meetings or threads could apply, ask **one** clarifying question — which meeting, which thread, which date range.
+1. **Bind** the engagement (`fde resume` / registry). If multiple meetings or threads could apply, ask **one** clarifying question - which meeting, which thread, which date range.
 2. **Capability check** (above). Then **fetch** via available source MCP(s). You pull; the CLI does not reach the network.
-3. **Stage** — `fde ingest stage [--source NAME] [--title TEXT] [file|-]` writes raw text into `<engagement>/.inbox/` (outside the memory git ledger).
-4. **List** (optional) — `fde ingest list` shows staged items when you need an id or filename.
-5. **Propose** — `fde ingest propose <id-or-filename>` runs the debrief `--smart` path on the staged body (+ provenance line). Opens `.debrief-propose`.
-6. **Rewrite prefixes** — same as debrief: lines without `decision:` / `risk:` / `delivery:` / `contact:` / `next:` need **you** to rewrite before showing the FDE. `--smart` is a gate, not a brain.
+3. **Stage** - `fde ingest stage [--source NAME] [--title TEXT] [file|-]` writes raw text into `<engagement>/.inbox/` (outside the memory git ledger).
+4. **List** (optional) - `fde ingest list` shows staged items when you need an id or filename.
+5. **Propose** - `fde ingest propose <id-or-filename>` runs the debrief `--smart` path on the staged body (+ provenance line). Opens `.debrief-propose`.
+6. **Rewrite prefixes** - same as debrief: lines without `decision:` / `risk:` / `delivery:` / `contact:` / `next:` need **you** to rewrite before showing the FDE. `--smart` is a gate, not a brain.
 7. **Show** the proposed routing in plain language. Wait for confirm.
-8. **Apply** — on FDE confirm only → `fde ingest apply` (= `fde debrief --apply`). On reject → stop; ask what to change.
+8. **Apply** - on FDE confirm only → `fde ingest apply` (= `fde debrief --apply`). On reject → stop; ask what to change.
 
 No invented names, meetings, or quotes. If the propose looks wrong, fix prefixes with judgment, then re-show before apply.
 
@@ -41,7 +41,7 @@ No invented names, meetings, or quotes. If the propose looks wrong, fix prefixes
 
 | Path | Role |
 |------|------|
-| `~/fde-engagements/<slug>/.inbox/` | Staging for raw pulls. Not the memory ledger. NDA surface — same home tree as `.fde/`. |
+| `~/fde-engagements/<slug>/.inbox/` | Staging for raw pulls. Not the memory ledger. NDA surface - same home tree as `.fde/`. |
 | `~/fde-engagements/<slug>/.fde/` | System of record (unchanged contract). |
 | `.fde/.debrief-propose` | Propose file (shared with debrief). |
 
@@ -56,15 +56,15 @@ fde ingest apply
 
 ## Provenance
 
-When a staged fact came from a named source, carry `via:<source>` on the applied line where useful (e.g. `via:granola`, `via:gmail`). Helps receipts and sponsor disputes later — not mandatory on every context line.
+When a staged fact came from a named source, carry `via:<source>` on the applied line where useful (e.g. `via:granola`, `via:gmail`). Helps receipts and sponsor disputes later - not mandatory on every context line.
 
 ## MCP sink + recipes
 
-Optional `mcp/fdeops-ingest` wraps the same verbs over stdio. Source MCPs remain separate — the FDE adds whichever fetch tools they trust. Setup coach: `ingest-connect.md`. Copy-paste recipes: `mcp/recipes/`.
+Optional `mcp/fdeops-ingest` wraps the same verbs over stdio. Source MCPs remain separate - the FDE adds whichever fetch tools they trust. Setup coach: `ingest-connect.md`. Copy-paste recipes: `mcp/recipes/`.
 
 ## Checkpoint
 
-Before apply, read back the 2–3 most consequential captures in one breath — same as debrief. Confirm which sources you staged and what would land in the record. Then stop.
+Before apply, read back the 2-3 most consequential captures in one breath - same as debrief. Confirm which sources you staged and what would land in the record. Then stop.
 
 ## Principles