@haaaiawd/loom 2.1.2 → 2.1.3

package/docs/PROMPT_CATALOG.md

@@ -1,116 +1,139 @@
-# LOOM Prompt and Message Catalog
-
-LOOM's prompts are cognitive interfaces, not a collection of role-playing personas. The exact live
-inventory is emitted by:
-
-```bash
-loom prompts
-```
-
-The command is the auditable source of truth. It returns the complete text, placeholders, and composition
-order as JSON, so documentation cannot quietly describe a different prompt from the one Agents receive.
-
-## Instruction layers
-
-| Layer | Message | Injected when | Purpose |
-|---|---|---|---|
-| Stable core | `layers.stable_core` | Every `loom context` | Durable collaboration judgment: user ownership, evidence classes, question threshold, corrections, reversible delegation |
-| Environment adaptation | `layers.environment_adaptation_available` or `layers.environment_adaptation_unavailable` | Every `loom context` | States whether a human can answer; routes uncertainty through inspection, permitted factual research, clarification, bounded assumption, or block |
-| Runtime adaptation | `layers.runtime_protocol` | Every `loom context` | LOOM-specific disk model, document graph, capability boundaries, Work Map, and three production loops |
-| Project state | `layers.project_state` | Every `loom context` | Dynamic status, uncertainty, assumptions, document/capability inventory, Task counts, and Keeper feedback |
-| Execution protocol | `layers.execution_protocol` | Active Task or `--task` | Recovery, workspace inspection, risk-based testing, progress persistence, exact proof, and delivery boundaries |
-| Current task | `layers.current_task` | Active Task or `--task` | Exact Task JSON followed by the exact contents of every path in `reads` |
-
-This separation is deliberate. Project-specific expertise does not inflate the stable prompt; it lives in
-field dossiers. One-off Task instructions do not become global doctrine. Host permissions and tools are not
-claimed by prompt text.
-
-When the human channel is unavailable, the environment message explicitly forbids fabricated user interaction and
-forbids treating web research as a substitute for user intent, preference, or authority. It preserves factual
-research when the task and host actually permit it, then requires a reversible assumption or block for what cannot
-be learned.
-
-## Workspace and document messages
-
-`workspace_anchor` is the short block added to `AGENTS.md`. It tells an entering or reset Agent to run
-`loom context`, rerun it before editing after an interruption, persist state at meaningful boundaries rather
-than before every tool call, and keep CLI operation invisible to the human.
-
-`templates.project_index` creates `.loom/PROJECT.md` as a concise whole and document map. It explicitly
-prevents a large project's systems from being compressed into one file.
-
-`templates.design_documents` contains all seven live design templates:
-
-1. `product` — outcome, users, problem, product principles, scope, end-to-end behavior, success and failure.
-2. `experience` — journey, information architecture, states, content, visual direction, accessibility, recovery, usability proof.
-3. `system` — responsibilities, boundaries, control flow, data, interfaces, dependencies, failure, implementation, verification.
-4. `contract` — consumers, schemas or commands, invariants, permissions, errors, compatibility, fixtures, contract tests.
-5. `verification` — claims, environments, fixtures, acceptance matrix, commands, negative tests, blind spots.
-6. `operations` — authority, procedures, commands, safety, failure recovery, evidence and audit trail.
-7. `research` — decision, evidence, method, sources, findings, conflicts, consequences and follow-up proof.
-
-These are possible document kinds, not a fixed seven-file checklist. The Agent creates as many actual
-documents as the project's independent decision surfaces require.
-
-`templates.professional_capability` is the only capability template. It enforces one recognizable field per
-file, a project-specific diagnosis and stance, opened evidence, rejected generic defaults, concrete design and
-verification consequences, failure modes, and links to adjacent fields without merging them.
-
-`templates.decision_history` is the preamble for consequential superseding decisions. Current truth stays in
-the project map and linked design documents.
-
-## Keeper message
-
-`keeper` is the exact independent handoff prompt. It binds a unique fresh-Agent `run_id` to a frozen
-`prepared_digest`, asks the Keeper to navigate the whole corpus and attempt the first Task, and defines three
-verdicts:
-
-- `passed`: a fresh Agent can responsibly begin.
-- `needs_revision`: concrete disk-truth gaps can be repaired.
-- `blocked`: progress needs unavailable authority or external state.
-
-A failed result is injected into the next project-state message. A retry requires changed project truth, a
-new digest, and another fresh Keeper.
-
-## Evil Eval messages
-
-`evaluation.baseline_condition` gives a normal capable Agent the identical brief and ordinary tools without
-LOOM. It does not weaken the baseline or prohibit normal planning.
-
-`evaluation.loom_condition` gives the same brief, model class, tools, workspace facts, and budget, adding only
-LOOM continuity infrastructure and no extra authority.
-
-`evaluation.unattended_baseline_condition` and `evaluation.unattended_loom_condition` are the same two
-conditions when no user response channel exists. The environment fact is equal in both arms; only the LOOM arm
-uses `loom context --human-channel unavailable`.
-
-`evaluation.blind_judge` scores anonymized output on intent fidelity, question value, whole coverage,
-professional depth, buildability, reset continuity, implementation evidence, human burden, and cost. It
-penalizes ceremony and runs an order swap.
-
-## Composition messages
-
-`composition.normal_resume`, `composition.active_task`, and `composition.keeper_handoff` list the exact block
-order. Keeper receives every design and capability document; a normal active Task receives only its declared
-reads. This difference is intentional: Keeper audits whole-project coverage, while delivery protects context
-focus.
-
-The active Task composition adds `execution_protocol` between the recovered project whole and the exact Task.
-It requires inspection of current workspace/version-control reality, risk-appropriate tests, restartable
-progress, and criterion-level proof. It does not force a unit test for non-code claims or a branch/PR for every
-Task; those are selected when the behavior, human request, or repository workflow makes them meaningful.
-
-## CLI operational messages
-
-`loom --help` is the complete command-oriented message surface. JSON command results report state and the next
-host action. Validation errors are contract enforcement rather than cognitive prompts; they reject unsafe
-references, template-only readiness, missing design documents, open high-impact uncertainty, unchanged Keeper
-retries, stale digests, duplicate run identities, Task dependency violations, missing or directory-level Task
-reads, status bypasses, completion without evidence for every done condition, and reasonless reopening of a
-disproven completion.
-
-The short human fallback when the host cannot create a Keeper Agent remains:
-
-```text
-Please open a new window in this project, run loom keeper prompt, and follow it.
-```
+# LOOM Prompt and Message Catalog
+
+LOOM's prompts are cognitive interfaces, not a collection of role-playing personas. The exact live
+inventory is emitted by:
+
+```bash
+loom prompts
+```
+
+The command is the auditable source of truth. It returns the complete text, placeholders, and composition
+order as JSON, so documentation cannot quietly describe a different prompt from the one Agents receive.
+
+## Instruction layers
+
+| Layer | Message | Injected when | Purpose |
+|---|---|---|---|
+| Stable core | `layers.stable_core` | Every `loom context` | Durable collaboration judgment: user ownership, evidence classes, question threshold, corrections, reversible delegation |
+| Environment adaptation | `layers.environment_adaptation_available` or `layers.environment_adaptation_unavailable` | Every `loom context` | States whether a human can answer; routes uncertainty through inspection, permitted factual research, clarification, bounded assumption, or block |
+| Runtime adaptation | `layers.runtime_protocol` | Every `loom context` | LOOM-specific disk model, document graph, capability boundaries, Work Map, and three production loops |
+| Project state | `layers.project_state` | Every `loom context` | Dynamic status, uncertainty, assumptions, document/capability inventory, Task counts, blocked Tasks with reasons and recovery conditions, and Keeper feedback |
+| Project state after failure | `layers.project_state_after_keeper_failure` | Keeper `needs_revision`/`blocked` | Prior finding text, the repair expectation, and the `loom project ready` → fresh-Keeper retry contract |
+| Keeper review of prior failure | `layers.keeper_review_of_prior_failure` | `--keeper` after a failed attempt | The exact prior findings the fresh Keeper must re-verify before passing |
+| Execution protocol | `layers.execution_protocol` | Active Task or `--task` | Recovery, workspace inspection, risk-based testing, progress persistence, exact proof, and delivery boundaries |
+| Current task | `layers.current_task` | Active Task or `--task` | Exact Task JSON followed by the exact contents of every path in `reads` |
+| On-demand map | `layers.on_demand_map` | Every `loom context` | Pointers to decision history, design documents, capability dossiers, and STRUCTURE.md for lazy loading |
+| Keeper context | `layers.keeper_context` | `--keeper` | Full decision history, work-map summary, and every design and capability document for whole-corpus audit |
+
+This separation is deliberate. Project-specific expertise does not inflate the stable prompt; it lives in
+field dossiers. One-off Task instructions do not become global doctrine. Host permissions and tools are not
+claimed by prompt text.
+
+When the human channel is unavailable, the environment message explicitly forbids fabricated user interaction and
+forbids treating web research as a substitute for user intent, preference, or authority. It preserves factual
+research when the task and host actually permit it, then requires a reversible assumption or block for what cannot
+be learned.
+
+## Workspace and document messages
+
+`workspace_anchor` is the short block added to `AGENTS.md`. It tells an entering or reset Agent to run
+`loom context`, rerun it before editing after an interruption, persist state at meaningful boundaries rather
+than before every tool call, and keep CLI operation invisible to the human. The anchor is durable-only:
+`loom init` never rewrites existing `AGENTS.md` content, and current mechanics always reach the Agent through
+`loom context` output rather than through the anchor text.
+
+`templates.project_index` creates `.loom/PROJECT.md` as a concise whole and document map. It explicitly
+prevents a large project's systems from being compressed into one file.
+
+`templates.design_documents` contains all seven live design templates:
+
+1. `product` — outcome, users, problem, product principles, scope, end-to-end behavior, success and failure.
+2. `experience` — journey, information architecture, states, content, visual direction, accessibility, recovery, usability proof.
+3. `system` — responsibilities, boundaries, control flow, data, interfaces, dependencies, failure, implementation, verification.
+4. `contract` — consumers, schemas or commands, invariants, permissions, errors, compatibility, fixtures, contract tests.
+5. `verification` — claims, environments, fixtures, acceptance matrix, commands, negative tests, blind spots.
+6. `operations` — authority, procedures, commands, safety, failure recovery, evidence and audit trail.
+7. `research` — decision, evidence, method, sources, findings, conflicts, consequences and follow-up proof.
+
+These are possible document kinds, not a fixed seven-file checklist. The Agent creates as many actual
+documents as the project's independent decision surfaces require.
+
+`templates.professional_capability` is the only capability template. It enforces one recognizable field per
+file, a project-specific diagnosis and stance, opened evidence, rejected generic defaults, concrete design and
+verification consequences, failure modes, and links to adjacent fields without merging them.
+
+`templates.decision_history` is the preamble for consequential superseding decisions. Current truth stays in
+the project map and linked design documents.
+
+## Keeper message
+
+`keeper` is the exact independent handoff prompt. It binds a unique fresh-Agent `run_id` to a frozen
+`prepared_digest`, asks the Keeper to navigate the whole corpus and attempt the first Task, and defines three
+verdicts:
+
+- `passed`: a fresh Agent can responsibly begin.
+- `needs_revision`: concrete disk-truth gaps can be repaired.
+- `blocked`: progress needs unavailable authority or external state.
+
+A `passed` verdict must carry independent review provenance (`review.mode: "independent"` plus a
+`reviewer_id` and evidence that a separate Agent ran the review). A self-review or an unattested pass is
+rejected at record time; a legacy unattested pass blocks `task start` and is an error in `loom check` until
+replaced by a fresh review or an explicit `loom keeper skip --reason`.
+
+Findings are cumulative across attempts: a pass must carry `closure_results` with evidence for every gap
+still open from earlier rounds, not only the latest attempt's. A changed project digest proves that
+something changed — never that findings were resolved. There is no auto-pass.
+
+A failed result is injected into the next project-state message. A retry requires changed project truth, a
+new digest, and another fresh Keeper. `loom review --help` describes the staging steps for opening that
+fresh Agent without inherited conversation.
+
+## Evil Eval messages
+
+`evaluation.baseline_condition` gives a normal capable Agent the identical brief and ordinary tools without
+LOOM. It does not weaken the baseline or prohibit normal planning.
+
+`evaluation.loom_condition` gives the same brief, model class, tools, workspace facts, and budget, adding only
+LOOM continuity infrastructure and no extra authority.
+
+`evaluation.unattended_baseline_condition` and `evaluation.unattended_loom_condition` are the same two
+conditions when no user response channel exists. The environment fact is equal in both arms; only the LOOM arm
+uses `loom context --human-channel unavailable`.
+
+`evaluation.blind_judge` scores anonymized output on intent fidelity, question value, whole coverage,
+professional depth, buildability, reset continuity, implementation evidence, human burden, and cost. It
+penalizes ceremony and runs an order swap.
+
+## Composition messages
+
+`composition.normal_resume`, `composition.active_task`, and `composition.keeper_handoff` list the exact block
+order. Keeper receives every design and capability document; a normal active Task receives only its declared
+reads. This difference is intentional: Keeper audits whole-project coverage, while delivery protects context
+focus.
+
+The active Task composition adds `execution_protocol` between the recovered project whole and the exact Task.
+It requires inspection of current workspace/version-control reality, risk-appropriate tests, restartable
+progress, and criterion-level proof. It does not force a unit test for non-code claims or a branch/PR for every
+Task; those are selected when the behavior, human request, or repository workflow makes them meaningful.
+
+## CLI operational messages
+
+`loom --help` is the complete command-oriented message surface. `loom review` and `loom review --help` are
+the read-only entry points an Agent uses to stage an independent Keeper handoff; `loom keeper --help` covers
+`keeper prompt`, `keeper record`, and `keeper skip`. JSON command results report state and the next
+host action. Validation errors are contract enforcement rather than cognitive prompts; they reject unsafe
+references, template-only readiness, missing design documents, open high-impact uncertainty, unchanged Keeper
+retries, stale digests, duplicate run identities, Task dependency violations, missing or directory-level Task
+reads, Task references to nonexistent design sections or capability nodes, status bypasses, completion
+without evidence for every done condition, silent rewriting of a done Task, and reasonless reopening of a
+disproven completion or unmet block.
+
+Warning channels are closable rather than permanent: a decision-affected done Task warns only until it is
+reopened and re-completed after the decision, deliverable coverage distinguishes planned from delivered, and
+blocked Tasks surface their recovery conditions in `loom context` until reopened with a concrete reason.
+
+The short human fallback when the host cannot create a Keeper Agent remains:
+
+```text
+Please open a new window in this project, run loom keeper prompt, and follow it.
+```