danielmiessler · simonasrazm · Jun 27, 2026 · Jun 27, 2026 · Jun 27, 2026 · Jun 27, 2026
diff --git a/Releases/v5.0.0/.claude/PAI/ALGORITHM/capabilities.md b/Releases/v5.0.0/.claude/PAI/ALGORITHM/capabilities.md
@@ -13,7 +13,7 @@ Use these to enrich understanding BEFORE or DURING ISC writing. Select in the pr
 | IterativeDepth | OBSERVE | **Default at Extended+** when time budget allows deeper understanding; any important task where exploring the full problem space before ISC improves outcome; understanding what's actually being asked vs what was literally said; exploring different approach angles before committing; ambiguous scope, multi-faceted problems, hidden assumptions | `Skill("IterativeDepth")` | E2+ |
 | ApertureOscillation | OBSERVE, THINK | Building something specific within a larger system; architecture decisions where scope framing changes the answer; feature design where tactical and strategic views may diverge; system coherence checks; scope negotiation. Complementary to IterativeDepth — ID rotates lenses, AO oscillates scope. Use AO when two distinct zoom levels (tactical target + strategic context) exist. | `Skill("ApertureOscillation")` | E3+ |
 | FeedbackMemoryConsult | PLAN | **First step of PLAN at Extended+.** Before committing to approach, grep `~/.claude/projects/${HARNESS_USER_DIR}/memory/feedback_*.md` by task keywords. Prevents repeating mistakes already documented. Turns the memory system from write-only diary into active guardrail. | `Bash('rg -l "KEYWORDS" ~/.claude/projects/${HARNESS_USER_DIR}/memory/feedback_*.md')` | E2+ |
-| Advisor | VERIFY | **At commitment boundaries on multi-step ISAs.** Before approach commitment, when stuck, once after durable deliverable before declaring done. Skip for short reactive tasks. If empirical results contradict advisor, re-call surfacing the conflict — do NOT silently switch. | `bun ~/.claude/PAI/TOOLS/Inference.ts --mode advisor <task> <state> <question>` | E3+ |
+| Advisor | VERIFY | **At commitment boundaries on multi-step ISAs.** Before approach commitment, when stuck, once after durable deliverable before declaring done. Skip for short reactive tasks. If empirical results contradict advisor, re-call surfacing the conflict — do NOT silently switch. | `bun ${PAI_DIR}/TOOLS/Inference.ts --mode advisor <task> <state> <question>` | E3+ |
 | ReReadCheck | VERIFY→LEARN boundary | **Final gate before emitting response (v3.29 RR1).** Re-read user's last message verbatim; enumerate every explicit ask against what shipped; block `phase: complete` on any `✗`. Targets the 82% "missed ask" complaint cluster. MANDATORY at every tier — at E1 single-part it's a one-line block. No fast-path exemption. | *(inline doctrine step — no external tool)* | E1+ |
 | FirstPrinciples | THINK | Architecture decisions, inherited assumptions, stuck on approach | `Skill("FirstPrinciples")` | E2+ |
 | SystemsThinking | OBSERVE, THINK | Recurring problems, structural causes, feedback loops, unintended consequences, "why does this keep happening?" Iceberg model, causal loop diagrams, Senge archetypes, Meadows' 12 leverage points | `Skill("SystemsThinking")` | E3+ |

diff --git a/Releases/v5.0.0/.claude/PAI/ALGORITHM/ideate-loop.md b/Releases/v5.0.0/.claude/PAI/ALGORITHM/ideate-loop.md
@@ -8,7 +8,7 @@ User says "ideate [problem]", "id8 [problem]", "generate ideas for [problem]", o
 
 ### Parameter Integration
 
-Parameters control ideation behavior along a spectrum from pure free-form dreaming to tight analytical focus. Full schema: `~/.claude/PAI/ALGORITHM/parameter-schema.md`.
+Parameters control ideation behavior along a spectrum from pure free-form dreaming to tight analytical focus. Full schema: `${PAI_DIR}/ALGORITHM/parameter-schema.md`.
 
 **Parameter resolution** (during OBSERVE phase):
 1. Algorithm detects preset/focus/params from user request

diff --git a/Releases/v5.0.0/.claude/PAI/ALGORITHM/mode-detection.md b/Releases/v5.0.0/.claude/PAI/ALGORITHM/mode-detection.md
@@ -26,7 +26,7 @@ When detected:
 **Triggers:** `ideate [problem]` | `id8 [problem]` | `generate ideas for` | `dream up solutions for`
 
 1. Set `mode: ideate` in ISA frontmatter
-2. Load `~/.claude/PAI/ALGORITHM/ideate-loop.md`
+2. Load `${PAI_DIR}/ALGORITHM/ideate-loop.md`
 3. Map effort tier to `time_scale` per ideate-loop.md
 
 ## Optimize Mode

diff --git a/Releases/v5.0.0/.claude/PAI/ALGORITHM/optimize-loop.md b/Releases/v5.0.0/.claude/PAI/ALGORITHM/optimize-loop.md
@@ -17,7 +17,7 @@ Referenced from the Algorithm when `mode: optimize`. This file defines the compl
 
 ### Parameter Integration
 
-Optimize mode accepts tunable parameters that control mutation boldness, regression tolerance, and patience. Full schema: `~/.claude/PAI/ALGORITHM/parameter-schema.md`.
+Optimize mode accepts tunable parameters that control mutation boldness, regression tolerance, and patience. Full schema: `${PAI_DIR}/ALGORITHM/parameter-schema.md`.
 
 **Optimize parameters:**
 

diff --git a/Releases/v5.0.0/.claude/PAI/ALGORITHM/v6.3.0.md b/Releases/v5.0.0/.claude/PAI/ALGORITHM/v6.3.0.md
@@ -12,7 +12,7 @@
 
 **The ISA has twelve sections (NEW v6.2.0).** Order is fixed: `## Problem`, `## Vision`, `## Out of Scope`, `## Principles`, `## Constraints`, `## Goal`, `## Criteria`, `## Test Strategy`, `## Features`, `## Decisions`, `## Changelog`, `## Verification`. Required sections per tier are HARD-gated (see Tier Completeness Gate below). Empty sections never appear — Bitter Pill discipline preserved. Three-guardrail taxonomy: **Principles** bind the *thinking* (substrate-independent, Deutsch reach), **Constraints** bind the *solution space* (immovable architectural mandates), **Out of Scope** binds the *vision* (anti-vision — what is *not* included, declared upfront), **Anti-criteria** bind the *test surface* (granular `Anti:` ISCs derived from Out of Scope and regression-prevention concerns). The first three are author-stated; anti-criteria are derived probes.
 
-**The ISA Skill (NEW v6.2.0)** at `~/.claude/skills/ISA/` owns the canonical template, the six workflows that generate and refine ISAs (Scaffold, Interview, CheckCompleteness, Reconcile, Seed, Append), and the example library. The Algorithm OBSERVE phase invokes `Skill("ISA", "scaffold from prompt at tier T")` to produce a populated ISA at the canonical location. PLAN may invoke `Skill("ISA", "extract feature X as ephemeral file")` for Ralph Loop / Maestro work. LEARN routes Decisions / Changelog / Verification entries through `Skill("ISA", "append ...")` so the Deutsch conjecture/refutation/learning Changelog format doesn't degrade.
+**The ISA Skill (NEW v6.2.0)** at `<harness-home>/skills/ISA/` owns the canonical template, the six workflows that generate and refine ISAs (Scaffold, Interview, CheckCompleteness, Reconcile, Seed, Append), and the example library. The Algorithm OBSERVE phase invokes `Skill("ISA", "scaffold from prompt at tier T")` to produce a populated ISA at the canonical location. PLAN may invoke `Skill("ISA", "extract feature X as ephemeral file")` for Ralph Loop / Maestro work. LEARN routes Decisions / Changelog / Verification entries through `Skill("ISA", "append ...")` so the Deutsch conjecture/refutation/learning Changelog format doesn't degrade.
 
 **The ISA is a living articulation.** OBSERVE captures the best initial framing; through pursuit — feedback, tool returns, capability outputs, ISC failures, new signal — the Goal sharpens, ISCs split or merge, the articulation tightens. Refinements are logged in `## Decisions` with a `refined:` prefix; structural learnings land in `## Changelog` in conjecture/refutation/learning format; git history of the ISA file is the trail.
 
@@ -79,13 +79,13 @@ REASON: <one sentence>
 SOURCE: classifier | fail-safe
 ```
 
-The executor reads this directly. **No regex fallback. No model judgment.** If `MODE` is MINIMAL or NATIVE, the executor uses the corresponding format from CLAUDE.md and stops. If `MODE` is ALGORITHM, the executor enters the Algorithm at the named `TIER`.
+The executor reads this directly. **No regex fallback. No model judgment.** If `MODE` is MINIMAL or NATIVE, the executor uses the corresponding format from the harness startup instructions and stops. If `MODE` is ALGORITHM, the executor enters the Algorithm at the named `TIER`.
 
 **Classifier rules (encoded in the hook's system prompt):**
 
 - **MINIMAL** — greetings, ratings, single-token acknowledgments.
 - **NATIVE** — single fact lookup OR single-line edit on a named file OR one command run, AND no new artifact created, AND no multi-step plan.
-- **ALGORITHM** — everything else. Includes any build/create/make/implement/design/refactor/migrate/integrate request, anything touching multiple files, anything ambiguous, anything affecting doctrine/system-prompt/hooks/CLAUDE.md/Algorithm/ISA, anything spanning multiple projects, any meta-question about the system itself.
+- **ALGORITHM** — everything else. Includes any build/create/make/implement/design/refactor/migrate/integrate request, anything touching multiple files, anything ambiguous, anything affecting doctrine/system-prompt/hooks/startup instructions/Algorithm/ISA, anything spanning multiple projects, any meta-question about the system itself.
 - **Tier (ALGORITHM only)** — E1 trivial (~<90s), E2 single-domain (~3min), E3 multi-file substantial (~10min), E4 cross-cutting/doctrine/architecture (~30min), E5 comprehensive (>2h). Bias higher when in doubt.
 
 **Override hierarchy (executor side):**
@@ -167,7 +167,7 @@ The `CheckCompleteness` workflow enforces this gate. A miss blocks `phase: compl
 - EXECUTE / VERIFY / LEARN: `Skill("ISA", "append <type> to <isa-path>: <content>")` — canonical writer for Decisions / Changelog / Verification.
 - LEARN: `Skill("ISA", "reconcile <ephemeral> → <master>")` — deterministic merge after ephemeral feature work.
 
-**v6.2.x deferred:** parser updates so `ISASync.hook.ts`, `CheckpointPerISC.hook.ts`, and `hooks/lib/isa-utils.ts` automatically discover `<project>/ISA.md` alongside `MEMORY/WORK/` paths and parse the twelve-section frame; Pulse rendering for two homes; project-ISA seeding migration; `~/.claude/skills/ISA/Tools/*.ts` CLI implementations. Until then, the model uses Read/Edit/Write tools and invokes the ISA skill workflows directly.
+**v6.2.x deferred:** parser updates so `ISASync.hook.ts`, `CheckpointPerISC.hook.ts`, and `hooks/lib/isa-utils.ts` automatically discover `<project>/ISA.md` alongside `MEMORY/WORK/` paths and parse the twelve-section frame; Pulse rendering for two homes; project-ISA seeding migration; `<harness-home>/skills/ISA/Tools/*.ts` CLI implementations. Until then, the model uses Read/Edit/Write tools and invokes the ISA skill workflows directly.
 
 ### ISC Quality System
 
@@ -224,7 +224,7 @@ Modes (ideate, optimize) accept tunable parameters. Full schema and presets: `PA
 
 **ALL WORK INSIDE THE ALGORITHM.** Every tool call, investigation, and decision happens within phases.
 
-**Entry banner was already printed by CLAUDE.md.** The user has seen:
+**Entry banner was already printed by the harness startup instructions.** The user has seen:
 ```
 ♻︎ Entering the PAI ALGORITHM… (v6.2.0) ═════════════
 🗒️ TASK: [8 word description]
@@ -235,7 +235,7 @@ Modes (ideate, optimize) accept tunable parameters. Full schema and presets: `PA
 **ISA stub** (immediately after voice):
 1. Determine ISA home: project ISA at `<project>/ISA.md` if task targets existing project; task ISA at `MEMORY/WORK/{slug}/ISA.md` for ad-hoc work
 2. **Invoke `Skill("ISA", "scaffold from prompt: <user message> at tier <tier>")`** — returns the populated ISA at canonical location with required sections per tier (NEW v6.2.0; replaces inline ISA construction)
-3. For task ISAs the skill creates `~/.claude/PAI/MEMORY/WORK/{slug}/`; for project ISAs the skill reads existing `<project>/ISA.md` if present, or seeds it via the Seed workflow
+3. For task ISAs the skill creates `${PAI_DIR}/MEMORY/WORK/{slug}/`; for project ISAs the skill reads existing `<project>/ISA.md` if present, or seeds it via the Seed workflow
 4. Skill output is the path; Algorithm reads/edits it via Read/Edit tools through subsequent phases
 
 **E1 fast-path exception:** at E1, the Algorithm may inline-write the minimal Goal+Criteria ISA without invoking the skill, to preserve the <90s budget. The skill invocation is mandatory at E2+.
@@ -347,7 +347,7 @@ Anti-criteria ≥1 and Antecedent ≥1-when-experiential are required. ID-stabil
 **Knowledge check (on-demand):** If the task topic has likely prior work, search `MEMORY/KNOWLEDGE/` for relevant notes.
 
 ```bash
-rg -i "TOPIC" ~/.claude/PAI/MEMORY/KNOWLEDGE/ --type md -l
+rg -i "TOPIC" ${PAI_DIR}/MEMORY/KNOWLEDGE/ --type md -l
 ```
 
 ```
@@ -467,7 +467,7 @@ Use `Skill("ISA", "append verification to <isa-path>: ISC-N <probe-type> <eviden
 
 ### 🪢 CHECKPOINTS (per-step durability)
 
-Every `[ ]`→`[x]` ISC transition fires `CheckpointPerISC.hook.ts`. For each repo in `~/.claude/checkpoint-repos.txt` with uncommitted changes, the hook auto-commits. Idempotent via sidecar `MEMORY/WORK/{slug}/.checkpoint-state.json`. (v6.2.x: hook will also fire on `<project>/ISA.md` edits once isa-utils discovery lands.)
+Every `[ ]`→`[x]` ISC transition fires `CheckpointPerISC.hook.ts`. For each repo in `<harness-home>/checkpoint-repos.txt` with uncommitted changes, the hook auto-commits. Idempotent via sidecar `MEMORY/WORK/{slug}/.checkpoint-state.json`. (v6.2.x: hook will also fire on `<project>/ISA.md` edits once isa-utils discovery lands.)
 
 ━━━ ✅ VERIFY ━━━ 6/7
 
@@ -503,7 +503,7 @@ On **multi-step ISAs** (Extended+ effort, multi-file edits, architecture changes
 3. **Once after producing a durable deliverable** — before setting `phase: complete` in LEARN
 
 ```bash
-bun ~/.claude/PAI/TOOLS/Inference.ts --mode advisor --auto-state \
+bun ${PAI_DIR}/TOOLS/Inference.ts --mode advisor --auto-state \
   "TASK: one-sentence description" \
   "QUESTION: specific decision point or 'any gaps before declaring done?'"
 ```
@@ -597,7 +597,7 @@ Agent({
 | TYPE | Target surface | Gate |
 |------|----------------|------|
 | `knowledge` | `MEMORY/KNOWLEDGE/{People\|Companies\|Ideas\|Research}/<slug>.md` | **Inline write.** |
-| `rule` | `CLAUDE.md` Operational Rules section | **Inline append.** |
+| `rule` | Harness startup instructions Operational Rules section | **Inline append.** |
 | `gotcha` | The relevant skill's `SKILL.md` Gotchas section | **Inline append.** |
 | `state` | `USER/PROJECTS/PROJECTS.md` "Open Sessions to Resume" | **Inline append.** |
 | `business` | `USER/BUSINESS/<topic>.md` | **Inline write/append.** |
@@ -629,7 +629,7 @@ Agent({
 
 **WRITE REFLECTION JSONL** (Extended+ effort; skipped at E1):
 ```bash
-echo '{"timestamp":"[ISO-8601]","effort_level":"[tier]","effort_source":"[auto|gate-floor|explicit]","task_description":"[TASK line]","criteria_count":[N],"criteria_passed":[N],"criteria_failed":[N],"prd_id":"[slug]","implied_sentiment":[1-10],"satisfaction_prediction":[1-10],"reflection_q1":"[Q1]","reflection_q2":"[Q2]","reflection_q3":"[Q3]","knowledge_flags":[N],"within_budget":[bool],"living_doc_refinements":[N],"doctrine_fired":{"live_probe":[bool],"advisor":[bool],"cato":[bool],"conflict":[bool],"thinking_floor_met":[bool],"completeness_gate_met":[bool]}}' >> ~/.claude/PAI/MEMORY/LEARNING/REFLECTIONS/algorithm-reflections.jsonl
+echo '{"timestamp":"[ISO-8601]","effort_level":"[tier]","effort_source":"[auto|gate-floor|explicit]","task_description":"[TASK line]","criteria_count":[N],"criteria_passed":[N],"criteria_failed":[N],"prd_id":"[slug]","implied_sentiment":[1-10],"satisfaction_prediction":[1-10],"reflection_q1":"[Q1]","reflection_q2":"[Q2]","reflection_q3":"[Q3]","knowledge_flags":[N],"within_budget":[bool],"living_doc_refinements":[N],"doctrine_fired":{"live_probe":[bool],"advisor":[bool],"cato":[bool],"conflict":[bool],"thinking_floor_met":[bool],"completeness_gate_met":[bool]}}' >> ${PAI_DIR}/MEMORY/LEARNING/REFLECTIONS/algorithm-reflections.jsonl
 ```
 
 ---
@@ -659,8 +659,8 @@ If after compaction you don't know your state:
 
 **Cold-start recovery (new session on existing work):**
 1. For project work: read `<project>/ISA.md`
-2. For task work: read ISA from `~/.claude/PAI/MEMORY/WORK/`
-3. `~/.claude/PAI/MEMORY/STATE/work.json` has the session registry
+2. For task work: read ISA from `${PAI_DIR}/MEMORY/WORK/`
+3. `${PAI_DIR}/MEMORY/STATE/work.json` has the session registry
 
 ---