@spences10/pi-harness 0.1.2 → 0.2.0

package/README.md

@@ -11,18 +11,19 @@
 
 <!-- package-readme:header:end -->
 
-Build and run ephemeral task harnesses with my-pi primitives. A
-harness is a `/tmp` runtime containing a machine-readable contract,
-executor prompt, task brief, validation script, review script, logs,
-status, and runtime enforcement. Inspired by Ornith's self-scaffolding
-design, the runtime separates an immutable outer policy from a
-versioned, amendable inner scaffold.
+Prompt guidance uses Pi's named system-prompt sections. It coexists
+with user `SYSTEM.md` and `APPEND_SYSTEM.md` files and other
+extensions without replacing the complete prompt.
 
-The harness tools use normal object-schema tool calling because their
-optional policy fields are not portable across providers'
-strict-schema rules.
+Optional task scaffolding for Pi: a contract, working-state baseline,
+validation scripts, progress log, and reviewable outcome. Pi still
+owns the agent loop, sessions, tools, and permission hooks.
 
-## Installation
+Use direct work for routine changes. Use a harness when the user asks
+for one or a concrete execution contract adds value. Assessment is
+optional advice, not a prerequisite or a read-only security mode.
+
+## Install
 
 <!-- package-readme:install:start -->
 
@@ -32,181 +33,125 @@ pi install npm:@spences10/pi-harness
 
 <!-- package-readme:install:end -->
 
-Local development from this monorepo:
+## Create and execute
 
-```bash
-pnpm --filter @spences10/pi-harness run build
-pi install ./packages/pi-harness
-# or for one run only
-pi -e ./packages/pi-harness
+```text
+/harness create <task>
 ```
 
-## What it does
-
-`pi-harness` adds:
-
-- `/assess` command for evidence-gated capability assessment
-- `/harness` command for assess-before-create, run, review, status,
-  use, and clear
-- `harness_assess` and `harness_assessment_submit` tools for read-only
-  orientation and structured adoption decisions
-- guarded `harness_create` tool that cannot bypass assessment approval
-- `harness_amend` tool for audited, versioned changes to the inner
-  scaffold
-- `harness_update` tool for phase/status/evidence logging
-- generated `OUTCOME.md` and `outcome.json` review artifacts
-- dirty-baseline tracking so pre-existing repo changes are reported
-  but not treated as task drift
-- worktree-aware validate/review scripts for team-mode executors
-- `harness_read` tool for summaries
-- `before_agent_start` context injection for the active harness
-- `tool_call` enforcement for edit/write paths and forbidden commands
-- compact TUI status for the active harness
-- bundled `create-harness`, `execute-harness`, and `review-harness`
-  skills
-
-## Runtime layout
+The agent calls `harness_create` with the task and at least one
+accepted validation command. Scope defaults to the project; provide
+`allowed_paths` to narrow it. Normal project test development is
+allowed by default. Set `allow_test_changes: false` only when the task
+explicitly requires protecting those files.
 
-```text
-/tmp/my-pi-harness-<id>/
-  harness.json      # outer policy, versioned inner scaffold, amendment history
-  SYSTEM.md         # executor system prompt
-  TASK.md           # task brief and required loop
-  status.json       # phase/status/evidence log
-  OUTCOME.md        # reviewable outcome summary
-  outcome.json      # machine-readable outcome summary
-  outcome.mjs       # outcome artifact refresher
-  validate.sh       # validation runner
-  review.sh         # drift/review helper
-  logs/events.jsonl # append-only event log
-```
+Creation does not require a separate assessment or approval roundtrip.
+It does not authorize new capabilities, commits, pushes, releases,
+deployments, destructive operations, or peer collaboration beyond the
+user's request.
+
+The runtime lives under `/tmp/my-pi-harness-<id>/`:
+
+- `harness.json`: accepted maximum scope and checks, plus the
+  execution scaffold and amendment history.
+- `TASK.md` and `SYSTEM.md`: task and workflow guidance.
+- `validate.sh` and `review.sh`: manual validation and drift review.
+- `status.json` and `logs/`: progress, executed checks and output.
+- `OUTCOME.md` and `outcome.json`: review artifacts.
+
+### Completion means checks ran
+
+Request `status: completed` through `harness_update`. The extension
+runs the accepted commands, any additional scaffold checks, and the
+drift guard through Pi's normal nested Bash tool pipeline. Permission
+hooks, cancellation and tool-result handling still apply.
+
+Completion requires successful checks against an unchanged contract
+and Git-visible working state. A failure, denied tool, interruption or
+concurrent change cannot become success merely because an agent writes
+“passed”. Results include executed-check records and output-file
+references. Manual evidence strings are self-reported notes, not
+validation proof.
+
+Failed checks can be fixed and retried; there is no one-attempt
+escalation rule. Completed runs reject further API amendments and
+status mutations. Their active guard clears on the next direct user
+turn or session startup; do not create another harness just to commit
+already-reviewed work.
+
+### Amend strategy, not acceptance
+
+`harness_amend` records a versioned strategy change. It cannot remove
+or replace the accepted validation commands or relax an explicit
+no-test-edit restriction. Additional checks are allowed. Scaffold
+scope always intersects the accepted maximum scope, so broadening a
+strategy does not grant access outside the contract.
+
+A genuinely larger task needs user authorization and a new contract,
+not an executor-authored relaxation of its own acceptance criteria.
+
+### Baseline and worktrees
 
-## Commands
+Dirty files are fingerprinted without changing the working tree, HEAD
+or Git index. Later content, index, mode, symlink, deletion and rename
+changes remain visible, including initially untracked files. Only
+fingerprints are saved, not file contents.
+
+A baseline belongs to the workspace where it was captured. Linked
+worktrees use their own Git state, without another workspace's dirty
+file exemptions. Run manual scripts from the execution worktree or set
+`HARNESS_CWD`. Completion uses the current Pi execution directory and
+rejects unrelated repositories.
+
+### Optional assessment
+
+`/assess <task>`, `harness_assess` and `harness_assessment_submit`
+record evidence, existing primitives and a recommendation. Tools,
+native codemode, discovery and MCP remain available. This is advisory
+investigation: it does not enforce read-only access or add an approval
+queue. Normal user authorization still applies.
+
+## Commands and tools
 
 ```text
-/assess <task>
-/assess status
-/assess approve [harness|direct]
-/assess reject
-/assess clear
 /harness create <task>
-/harness run <dir>
-/harness review <dir>
+/harness run [dir]
+/harness review [dir]
 /harness status [dir]
 /harness use <dir>
 /harness clear
+/assess <task>
+/assess status
+/assess clear
 ```
 
-`/assess` and `/harness create` enter the same read-only assessment.
-The agent records repository evidence, existing primitives, rejected
-options, the smallest useful vertical slice, and one recommendation. A
-harness recommendation must include explicit allowed paths and
-validation commands. Direct user approval then creates and activates
-the harness, restores tools for approved direct work, or rejects the
-candidate without project changes. Approved harnesses run through the
-bundled `execute-harness` skill. Harness approval does not authorize
-Factory; Factory requires a separate direct user request for
-evaluation.
-
-An agent can enter assessment with `harness_assess`. If it calls
-`harness_create` without approval, the extension blocks the call and
-starts assessment. This backstop does not detect an agent that
-misclassifies a task and never asks for harness machinery.
-
-Use assessment when a task may benefit from an enforceable execution
-contract because it has material risk, unresolved scope, destructive
-effects, or complex coordination. Use the normal direct workflow for
-bounded, low-risk changes that standard validation can verify, such as
-documentation and copy edits, focused single-file fixes, configuration
-or metadata updates, test expectation changes, formatting, and
-reviewed commit or push follow-ups. A simple ambiguity may need only a
-clarifying question. Broad uncertain refactors, migrations,
-deployments, risky releases, external side effects, and explicit user
-requests are good assessment candidates.
-
-Assessment snapshots the current tool set, keeps an explicit read-only
-research subset active, and filters Bash through a strict
-single-command allowlist. Unknown and mutating custom tools are
-withheld. This is a workflow guard, not an operating-system sandbox.
-In print and JSON modes, an assessment can be submitted but remains
-waiting because direct approval cannot be inferred. RPC clients can
-answer Pi's extension UI request, or send `/assess approve ...` as an
-explicit command.
-
-`harness.json` contains two deliberately different layers:
-
-- `policy`: runtime-owned workspace and verifier protections such as
-  cwd, forbidden paths/commands, available tools, and the dirty
-  baseline. Executors cannot weaken these through amendments.
-- `scaffold`: the model's task interpretation and execution strategy,
-  including allowed paths, validation, test policy, and model roles.
-  It is subordinate to system, developer, and current user
-  instructions and can be revised with `harness_amend`.
-
-Every scaffold amendment increments its version, records its reason
-and requester, regenerates runtime prompts/scripts, and appends an
-audit event. A harness enforcement block reports a contract mismatch;
-it is not a platform-policy refusal. When a harness is active, the TUI
-shows a compact footer/status indicator; use `/harness status [dir]`
-or `harness_read` for the full contract, task, validation, and outcome
-details.
-
-A terminal `completed` or `failed` status seals the run and keeps
-enforcement active for the remainder of the current turn, so an
-executor cannot escape the outer policy by declaring its own work
-complete. The extension automatically deactivates that terminal
-harness on the next direct user turn or when the session starts again.
-Extension-injected messages do not trigger cleanup. Use
-`/harness clear` only to abandon a non-terminal run. Clearing or
-deleting a harness affects only its `/tmp` control state; repository
-changes remain in the working tree. Do not create a new harness merely
-to commit or push already-reviewed work. A risky release, deployment,
-migration, or destructive operation can still justify a dedicated
-harness when explicitly requested or warranted by its actual risk.
-
-Harnesses snapshot dirty files at creation time. Review and outcome
-artifacts focus on changes after that baseline, while still recording
-baseline files in `outcome.json`. If a team-mode executor runs in a
-linked git worktree, run `review.sh` from that worktree or set
-`HARNESS_CWD` so validation, guard checks, and outcome collection use
-the executor workspace.
-
-## Assessment and harness loops
-
-Assessment happens before harness creation:
-
-1. Read-only context recovery
-2. Source-of-truth and existing-primitive capture
-3. Assumption challenge and rejected options
-4. Smallest vertical slice
-5. Structured recommendation
-6. Direct user adoption decision
-
-An approved harness then runs:
-
-1. Contract recovery
-2. Surgical execution
-3. Validation evidence
-4. Drift review
-5. OUTCOME.md/outcome.json review
-6. Delta/risk report
-
-Use `harness_amend` for bounded inner-scaffold changes. A new
-capability or architecture decision requires assessment; an approved
-harness recommendation can then amend the active scaffold.
-Outer-policy expansion requires a separate execution contract instead
-of silent scope growth.
-
-## Using from a custom harness
-
-```ts
-import harness from '@spences10/pi-harness';
-
-// pass `harness` as an ExtensionFactory to your Pi runtime
-```
+Tools: `harness_create`, `harness_amend`, `harness_update`,
+`harness_read`, `harness_assess`, `harness_assessment_submit`.
+
+Execution is direct by default. Team Mode can coordinate independently
+opened peers only when the user authorizes collaboration. This package
+does not spawn or supervise workers.
+
+## Safety and compatibility
 
-`my-pi` imports this package directly and enables it as the built-in
-harness workflow.
+This is workflow scaffolding, **not an OS sandbox**. Direct edit/write
+hooks and final Git drift checks enforce the accepted paths through
+those interfaces. Shell commands and other tools retain the host
+user's permissions and can modify control files; the runtime is not a
+security boundary against hostile code. Use OS/container isolation for
+that threat model. Explicit `forbidden_commands` are optional
+substring checks, not a general shell parser.
+
+Existing temporary runtimes keep old generated scripts: recreate them
+before relying on these changes. Contracts without accepted checks and
+scope cannot complete or amend through the current API.
+
+For programmatic consumers, `create_harness_runtime` now requires
+`validation_commands`. `update_harness_runtime` is asynchronous;
+completion requires a trusted `HarnessVerificationOptions.run` backend
+that executes the supplied commands in the supplied cwd and respects
+the signal. The extension supplies this through Pi's tool pipeline.
+Removed assessment-locking helpers are no longer exported.
 
 ## Development
 
@@ -215,7 +160,3 @@ pnpm --filter @spences10/pi-harness run check
 pnpm --filter @spences10/pi-harness run test
 pnpm --filter @spences10/pi-harness run build
 ```
-
-## License
-
-MIT

package/dist/assessment.d.ts

@@ -16,7 +16,7 @@ export declare const harness_assessment_submit_params_schema: Type.TObject<{
         cwd: Type.TOptional<Type.TString>;
         allowed_paths: Type.TOptional<Type.TArray<Type.TString>>;
         forbidden_paths: Type.TOptional<Type.TArray<Type.TString>>;
-        validation_commands: Type.TOptional<Type.TArray<Type.TString>>;
+        validation_commands: Type.TArray<Type.TString>;
         forbidden_commands: Type.TOptional<Type.TArray<Type.TString>>;
         allow_test_changes: Type.TOptional<Type.TBoolean>;
         escalation_rules: Type.TOptional<Type.TArray<Type.TString>>;
@@ -32,7 +32,7 @@ export type HarnessAssessParams = Static<typeof harness_assess_params_schema>;
 export type HarnessAssessmentRecord = Static<typeof harness_assessment_submit_params_schema>;
 export type HarnessAssessmentRecommendation = HarnessAssessmentRecord['recommendation'];
 export type HarnessAssessmentSource = 'agent' | 'user' | 'harness-create';
-export type HarnessAssessmentStatus = 'assessing' | 'awaiting_approval' | 'approved' | 'rejected';
+export type HarnessAssessmentStatus = 'assessing' | 'assessed' | 'awaiting_approval' | 'approved' | 'rejected';
 export interface HarnessAssessmentState {
     id: string;
     version: number;
@@ -40,20 +40,12 @@ export interface HarnessAssessmentState {
     task: string;
     source: HarnessAssessmentSource;
     started_at: string;
-    tools_before_assessment: string[];
+    tools_before_assessment?: string[];
     record?: HarnessAssessmentRecord;
     decision?: HarnessAssessmentRecommendation;
 }
-export declare function create_assessment_state(task: string, source: HarnessAssessmentSource, tools_before_assessment: string[], now?: Date): HarnessAssessmentState;
+export declare function create_assessment_state(task: string, source: HarnessAssessmentSource, tools_before_assessment?: string[], now?: Date): HarnessAssessmentState;
 export declare function submit_assessment_record(state: HarnessAssessmentState, record: HarnessAssessmentRecord): HarnessAssessmentState;
 export declare function assessment_active(state: HarnessAssessmentState | undefined): boolean;
-export declare function assessment_tool_names(active_tool_names: string[]): string[];
-export declare function is_assessment_tool_allowed(tool_name: string): boolean;
-export declare function check_assessment_command(command: string): {
-    ok: true;
-} | {
-    ok: false;
-    reason: string;
-};
 export declare function assessment_context(state: HarnessAssessmentState): string;
 export declare function format_assessment_record(state: HarnessAssessmentState): string;

package/dist/assessment.js

@@ -3,25 +3,16 @@ import { harness_create_params_schema } from './schema.js';
 export const HARNESS_ASSESSMENT_CUSTOM_TYPE = 'pi-harness-assessment-state';
 export const harness_assess_params_schema = Type.Object({
     task: Type.String({
-        description: 'Candidate task to investigate before choosing an execution path',
+        minLength: 1,
+        description: 'Optional investigation before choosing an execution path',
     }),
 });
 export const harness_assessment_submit_params_schema = Type.Object({
-    task: Type.String({
-        description: 'Refined outcome being assessed',
-    }),
-    evidence: Type.Array(Type.String({
-        description: 'Concrete repository, documentation, or observed evidence',
-    })),
-    existing_primitives: Type.Array(Type.String({
-        description: 'Existing mechanisms that already satisfy part or all of the need',
-    })),
-    rejected_options: Type.Array(Type.String({
-        description: 'Rejected option and the evidence-based reason for rejection',
-    })),
-    smallest_vertical_slice: Type.String({
-        description: 'Smallest useful outcome or experiment that can test the need',
-    }),
+    task: Type.String(),
+    evidence: Type.Array(Type.String()),
+    existing_primitives: Type.Array(Type.String()),
+    rejected_options: Type.Array(Type.String()),
+    smallest_vertical_slice: Type.String(),
     recommendation: Type.Union([
         Type.Literal('harness'),
         Type.Literal('direct'),
@@ -29,204 +20,40 @@ export const harness_assessment_submit_params_schema = Type.Object({
     ]),
     proposed_contract: Type.Optional(harness_create_params_schema),
 });
-const ASSESSMENT_TOOL_NAMES = new Set([
-    'bash',
-    'context_get',
-    'context_list',
-    'context_search',
-    'context_stats',
-    'find',
-    'grep',
-    'harness_assess',
-    'harness_assessment_submit',
-    'harness_read',
-    'ls',
-    'lsp_definition',
-    'lsp_diagnostics',
-    'lsp_diagnostics_many',
-    'lsp_document_symbols',
-    'lsp_find_symbol',
-    'lsp_hover',
-    'lsp_references',
-    'mcp__mcp-omnisearch__ai_search',
-    'mcp__mcp-omnisearch__web_extract',
-    'mcp__mcp-omnisearch__web_search',
-    'mcp__mcp-sqlite-tools__close_database',
-    'mcp__mcp-sqlite-tools__database_info',
-    'mcp__mcp-sqlite-tools__describe_table',
-    'mcp__mcp-sqlite-tools__execute_read_query',
-    'mcp__mcp-sqlite-tools__export_schema',
-    'mcp__mcp-sqlite-tools__list_databases',
-    'mcp__mcp-sqlite-tools__list_tables',
-    'mcp__mcp-sqlite-tools__open_database',
-    'read',
-]);
-const SAFE_COMMAND_PATTERNS = [
-    /^(?:pwd|whoami|uname|date|uptime)(?:\s+.*)?$/,
-    /^(?:cat|head|tail|less|more|grep|rg|ls|wc|sort|uniq|diff|file|stat|du|df|tree|which|whereis|type|bat|eza)(?:\s+.*)?$/,
-    /^find(?:\s+.*)?$/,
-    /^git\s+(?:status|log|diff|show|rev-parse|ls-files|ls-tree|grep)(?:\s+.*)?$/,
-    /^git\s+branch(?:\s+(?:--list|-a|--all|-r|--remotes|--show-current|--contains|--merged|--no-merged)(?:\s+.*)?)?$/,
-    /^git\s+remote(?:\s+(?:-v|show|get-url)(?:\s+.*)?)?$/,
-    /^git\s+config\s+--get(?:\s+.*)?$/,
-    /^gh\s+(?:issue\s+(?:list|view)|pr\s+(?:list|view|diff|checks)|repo\s+view)(?:\s+.*)?$/,
-    /^gh\s+api(?:\s+.*)?$/,
-    /^(?:npm|pnpm|yarn)\s+(?:list|ls|view|info|why|outdated)(?:\s+.*)?$/,
-    /^(?:node|python|python3)\s+--version$/,
-    /^jq(?:\s+.*)?$/,
-    /^sed\s+-n(?:\s+.*)?$/,
-];
-const UNSAFE_READ_COMMAND_PATTERNS = [
-    /\bfind\b.*\s-(?:delete|exec|execdir|ok|okdir|fprint|fprintf|fls)\b/i,
-    /\brg\b.*\s--pre(?:-glob)?\b/i,
-    /\b(?:git\s+(?:log|diff|show)|sort)\b.*(?:\s--output(?:=|\s+)|\s-o\s+)/i,
-    /\bsed\b.*\s-i(?:\s|$|[^a-z])/i,
-    /\bgh\s+api\b.*(?:\s-X\s*|\s--method(?:=|\s+))(?!GET\b)/i,
-    /\bgh\s+api\b.*(?:\s-f\s|\s-F\s|\s--field(?:=|\s+)|\s--raw-field(?:=|\s+))/i,
-    /\bgit\s+branch\b.*\s-[dDmM]\b/i,
-];
-export function create_assessment_state(task, source, tools_before_assessment, now = new Date()) {
-    const normalized_task = task.trim();
-    if (!normalized_task) {
+export function create_assessment_state(task, source, tools_before_assessment = [], now = new Date()) {
+    if (!task.trim())
         throw new Error('Assessment task is required');
-    }
     return {
         id: `assessment-${now.getTime().toString(36)}`,
         version: 1,
         status: 'assessing',
-        task: normalized_task,
+        task: task.trim(),
         source,
         started_at: now.toISOString(),
-        tools_before_assessment: [...new Set(tools_before_assessment)],
+        tools_before_assessment,
     };
 }
 export function submit_assessment_record(state, record) {
-    if (record.recommendation === 'harness') {
-        if (!record.proposed_contract) {
-            throw new Error('A harness recommendation requires a proposed contract');
-        }
-        if (!record.proposed_contract.allowed_paths?.length) {
-            throw new Error('An approved harness contract requires explicit allowed paths');
-        }
-        if (!record.proposed_contract.validation_commands?.length) {
-            throw new Error('An approved harness contract requires validation commands');
-        }
-    }
     return {
         ...state,
         version: state.version + 1,
-        status: 'awaiting_approval',
+        status: 'assessed',
         task: record.task.trim(),
         record,
-        decision: undefined,
     };
 }
 export function assessment_active(state) {
-    return (state?.status === 'assessing' ||
-        state?.status === 'awaiting_approval');
-}
-export function assessment_tool_names(active_tool_names) {
-    const read_tools = [...new Set(active_tool_names)].filter((name) => ASSESSMENT_TOOL_NAMES.has(name));
-    return [
-        ...new Set([
-            ...read_tools,
-            'harness_assess',
-            'harness_assessment_submit',
-            'harness_read',
-        ]),
-    ];
-}
-export function is_assessment_tool_allowed(tool_name) {
-    return ASSESSMENT_TOOL_NAMES.has(tool_name);
-}
-function has_unquoted_shell_control(command) {
-    let quote;
-    for (let index = 0; index < command.length; index++) {
-        const character = command[index];
-        const next = command[index + 1];
-        if (quote === 'single') {
-            if (character === "'")
-                quote = undefined;
-            continue;
-        }
-        if (quote === 'double') {
-            if (character === '\\') {
-                index++;
-                continue;
-            }
-            if (character === '"') {
-                quote = undefined;
-                continue;
-            }
-            if (character === '`' || (character === '$' && next === '(')) {
-                return true;
-            }
-            continue;
-        }
-        if (character === '\\') {
-            index++;
-            continue;
-        }
-        if (character === "'") {
-            quote = 'single';
-            continue;
-        }
-        if (character === '"') {
-            quote = 'double';
-            continue;
-        }
-        if (character === '\n' ||
-            character === '\r' ||
-            character === ';' ||
-            character === '&' ||
-            character === '|' ||
-            character === '<' ||
-            character === '>' ||
-            character === '`' ||
-            (character === '$' && next === '(')) {
-            return true;
-        }
-    }
-    return quote !== undefined;
-}
-export function check_assessment_command(command) {
-    const normalized = command.trim();
-    if (!normalized) {
-        return { ok: false, reason: 'Assessment mode: empty command' };
-    }
-    if (has_unquoted_shell_control(normalized) ||
-        UNSAFE_READ_COMMAND_PATTERNS.some((pattern) => pattern.test(normalized))) {
-        return {
-            ok: false,
-            reason: 'Assessment mode: shell composition or a mutating command option is not allowed',
-        };
-    }
-    if (!SAFE_COMMAND_PATTERNS.some((pattern) => pattern.test(normalized))) {
-        return {
-            ok: false,
-            reason: 'Assessment mode: command is not in the read-only allowlist; use one read command per call',
-        };
-    }
-    return { ok: true };
+    return state?.status === 'assessing';
 }
 export function assessment_context(state) {
-    const approval = state.status === 'awaiting_approval'
-        ? 'An assessment record is waiting for direct user approval. Do not start implementation or call harness_create.'
-        : 'Investigate with read-only tools, then call harness_assessment_submit with one evidence-based recommendation.';
-    return `## Active harness assessment\n\nAssessment: \`${state.id}\` v${state.version}\nTask: ${state.task}\n\nRepository mutation is disabled. Treat proposed capabilities as candidates, not tasks. Establish the current source of truth, find existing primitives, reject unsupported options, and define the smallest useful vertical slice. A harness contract is optional and is valid only when the evidence shows that an enforceable execution contract adds value. Bash accepts one allowlisted read command per call; use the dedicated read, find, grep, LSP, and research tools when available.\n\n${approval}`;
+    return `## Optional harness assessment\n\nTask: ${state.task}\n\nInvestigate the need, existing primitives, and smallest useful change before recommending an execution path. This is advisory planning, not a read-only security mode: tools remain available and normal user authorization still applies. Submit findings with harness_assessment_submit. Assessment is not required to create a harness.`;
 }
 export function format_assessment_record(state) {
     const record = state.record;
-    if (!record) {
-        return `Assessment ${state.id} v${state.version}: ${state.status}\n\nTask: ${state.task}`;
-    }
-    const list = (values, empty) => values.length
-        ? values.map((value) => `- ${value}`).join('\n')
-        : `- ${empty}`;
-    const contract = record.proposed_contract;
-    const contract_section = contract
-        ? `\n\n## Proposed harness contract\n\n- Task: ${contract.task}\n- Cwd: ${contract.cwd ?? 'current project'}\n- Allowed paths: ${contract.allowed_paths?.join(', ') ?? 'none'}\n- Validation: ${contract.validation_commands?.join(' && ') ?? 'none'}\n- Test changes: ${contract.allow_test_changes ? 'allowed' : 'not allowed'}\n- Forbidden paths: ${contract.forbidden_paths?.join(', ') ?? 'defaults'}\n- Forbidden commands: ${contract.forbidden_commands?.join(', ') ?? 'defaults'}\n- Escalation rules: ${contract.escalation_rules?.join('; ') ?? 'defaults'}\n- Planner: ${contract.planner_model ?? 'default'} (${contract.planner_thinking ?? 'default'})\n- Executor: ${contract.executor_model ?? 'default'} (${contract.executor_thinking ?? 'default'})\n- Reviewer: ${contract.reviewer_model ?? 'default'} (${contract.reviewer_thinking ?? 'default'})`
-        : '';
-    return `# Harness assessment: ${state.id} v${state.version}\n\n- Status: ${state.status}\n- Recommendation: ${record.recommendation}\n\n## Outcome\n\n${record.task}\n\n## Evidence\n\n${list(record.evidence, 'No evidence recorded')}\n\n## Existing primitives\n\n${list(record.existing_primitives, 'None found')}\n\n## Rejected options\n\n${list(record.rejected_options, 'None recorded')}\n\n## Smallest vertical slice\n\n${record.smallest_vertical_slice}${contract_section}`;
+    if (!record)
+        return `Assessment ${state.id}: ${state.task}`;
+    const list = (values) => values.map((value) => `- ${value}`).join('\n') ||
+        '- None recorded';
+    return `# Harness assessment\n\n${record.task}\n\nRecommendation: ${record.recommendation}\n\n## Evidence\n\n${list(record.evidence)}\n\n## Existing primitives\n\n${list(record.existing_primitives)}\n\n## Rejected options\n\n${list(record.rejected_options)}\n\n## Smallest vertical slice\n\n${record.smallest_vertical_slice}`;
 }
 //# sourceMappingURL=assessment.js.map

package/dist/assessment.js.map

@@ -1 +1 @@
-{"version":3,"file":"assessment.js","sourceRoot":"","sources":["../src/assessment.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAe,MAAM,SAAS,CAAC;AAC5C,OAAO,EAAE,4BAA4B,EAAE,MAAM,aAAa,CAAC;AAE3D,MAAM,CAAC,MAAM,8BAA8B,GAC1C,6BAA6B,CAAC;AAE/B,MAAM,CAAC,MAAM,4BAA4B,GAAG,IAAI,CAAC,MAAM,CAAC;IACvD,IAAI,EAAE,IAAI,CAAC,MAAM,CAAC;QACjB,WAAW,EACV,iEAAiE;KAClE,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,uCAAuC,GAAG,IAAI,CAAC,MAAM,CAAC;IAClE,IAAI,EAAE,IAAI,CAAC,MAAM,CAAC;QACjB,WAAW,EAAE,gCAAgC;KAC7C,CAAC;IACF,QAAQ,EAAE,IAAI,CAAC,KAAK,CACnB,IAAI,CAAC,MAAM,CAAC;QACX,WAAW,EACV,0DAA0D;KAC3D,CAAC,CACF;IACD,mBAAmB,EAAE,IAAI,CAAC,KAAK,CAC9B,IAAI,CAAC,MAAM,CAAC;QACX,WAAW,EACV,kEAAkE;KACnE,CAAC,CACF;IACD,gBAAgB,EAAE,IAAI,CAAC,KAAK,CAC3B,IAAI,CAAC,MAAM,CAAC;QACX,WAAW,EACV,6DAA6D;KAC9D,CAAC,CACF;IACD,uBAAuB,EAAE,IAAI,CAAC,MAAM,CAAC;QACpC,WAAW,EACV,8DAA8D;KAC/D,CAAC;IACF,cAAc,EAAE,IAAI,CAAC,KAAK,CAAC;QAC1B,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC;QACvB,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC;QACtB,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC;KACtB,CAAC;IACF,iBAAiB,EAAE,IAAI,CAAC,QAAQ,CAAC,4BAA4B,CAAC;CAC9D,CAAC,CAAC;AAgCH,MAAM,qBAAqB,GAAG,IAAI,GAAG,CAAC;IACrC,MAAM;IACN,aAAa;IACb,cAAc;IACd,gBAAgB;IAChB,eAAe;IACf,MAAM;IACN,MAAM;IACN,gBAAgB;IAChB,2BAA2B;IAC3B,cAAc;IACd,IAAI;IACJ,gBAAgB;IAChB,iBAAiB;IACjB,sBAAsB;IACtB,sBAAsB;IACtB,iBAAiB;IACjB,WAAW;IACX,gBAAgB;IAChB,gCAAgC;IAChC,kCAAkC;IAClC,iCAAiC;IACjC,uCAAuC;IACvC,sCAAsC;IACtC,uCAAuC;IACvC,2CAA2C;IAC3C,sCAAsC;IACtC,uCAAuC;IACvC,oCAAoC;IACpC,sCAAsC;IACtC,MAAM;CACN,CAAC,CAAC;AAEH,MAAM,qBAAqB,GAAG;IAC7B,8CAA8C;IAC9C,sHAAsH;IACtH,kBAAkB;IAClB,4EAA4E;IAC5E,iHAAiH;IACjH,qDAAqD;IACrD,kCAAkC;IAClC,uFAAuF;IACvF,sBAAsB;IACtB,oEAAoE;IACpE,uCAAuC;IACvC,gBAAgB;IAChB,sBAAsB;CACtB,CAAC;AAEF,MAAM,4BAA4B,GAAG;IACpC,qEAAqE;IACrE,8BAA8B;IAC9B,wEAAwE;IACxE,+BAA+B;IAC/B,yDAAyD;IACzD,4EAA4E;IAC5E,gCAAgC;CAChC,CAAC;AAEF,MAAM,UAAU,uBAAuB,CACtC,IAAY,EACZ,MAA+B,EAC/B,uBAAiC,EACjC,GAAG,GAAG,IAAI,IAAI,EAAE;IAEhB,MAAM,eAAe,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IACpC,IAAI,CAAC,eAAe,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CAAC,6BAA6B,CAAC,CAAC;IAChD,CAAC;IACD,OAAO;QACN,EAAE,EAAE,cAAc,GAAG,CAAC,OAAO,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,EAAE;QAC9C,OAAO,EAAE,CAAC;QACV,MAAM,EAAE,WAAW;QACnB,IAAI,EAAE,eAAe;QACrB,MAAM;QACN,UAAU,EAAE,GAAG,CAAC,WAAW,EAAE;QAC7B,uBAAuB,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,uBAAuB,CAAC,CAAC;KAC9D,CAAC;AACH,CAAC;AAED,MAAM,UAAU,wBAAwB,CACvC,KAA6B,EAC7B,MAA+B;IAE/B,IAAI,MAAM,CAAC,cAAc,KAAK,SAAS,EAAE,CAAC;QACzC,IAAI,CAAC,MAAM,CAAC,iBAAiB,EAAE,CAAC;YAC/B,MAAM,IAAI,KAAK,CACd,uDAAuD,CACvD,CAAC;QACH,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,iBAAiB,CAAC,aAAa,EAAE,MAAM,EAAE,CAAC;YACrD,MAAM,IAAI,KAAK,CACd,8DAA8D,CAC9D,CAAC;QACH,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,iBAAiB,CAAC,mBAAmB,EAAE,MAAM,EAAE,CAAC;YAC3D,MAAM,IAAI,KAAK,CACd,2DAA2D,CAC3D,CAAC;QACH,CAAC;IACF,CAAC;IACD,OAAO;QACN,GAAG,KAAK;QACR,OAAO,EAAE,KAAK,CAAC,OAAO,GAAG,CAAC;QAC1B,MAAM,EAAE,mBAAmB;QAC3B,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE;QACxB,MAAM;QACN,QAAQ,EAAE,SAAS;KACnB,CAAC;AACH,CAAC;AAED,MAAM,UAAU,iBAAiB,CAChC,KAAyC;IAEzC,OAAO,CACN,KAAK,EAAE,MAAM,KAAK,WAAW;QAC7B,KAAK,EAAE,MAAM,KAAK,mBAAmB,CACrC,CAAC;AACH,CAAC;AAED,MAAM,UAAU,qBAAqB,CACpC,iBAA2B;IAE3B,MAAM,UAAU,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,iBAAiB,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAClE,qBAAqB,CAAC,GAAG,CAAC,IAAI,CAAC,CAC/B,CAAC;IACF,OAAO;QACN,GAAG,IAAI,GAAG,CAAC;YACV,GAAG,UAAU;YACb,gBAAgB;YAChB,2BAA2B;YAC3B,cAAc;SACd,CAAC;KACF,CAAC;AACH,CAAC;AAED,MAAM,UAAU,0BAA0B,CACzC,SAAiB;IAEjB,OAAO,qBAAqB,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;AAC7C,CAAC;AAED,SAAS,0BAA0B,CAAC,OAAe;IAClD,IAAI,KAAsC,CAAC;IAC3C,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,OAAO,CAAC,MAAM,EAAE,KAAK,EAAE,EAAE,CAAC;QACrD,MAAM,SAAS,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC;QACjC,MAAM,IAAI,GAAG,OAAO,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;QAChC,IAAI,KAAK,KAAK,QAAQ,EAAE,CAAC;YACxB,IAAI,SAAS,KAAK,GAAG;gBAAE,KAAK,GAAG,SAAS,CAAC;YACzC,SAAS;QACV,CAAC;QACD,IAAI,KAAK,KAAK,QAAQ,EAAE,CAAC;YACxB,IAAI,SAAS,KAAK,IAAI,EAAE,CAAC;gBACxB,KAAK,EAAE,CAAC;gBACR,SAAS;YACV,CAAC;YACD,IAAI,SAAS,KAAK,GAAG,EAAE,CAAC;gBACvB,KAAK,GAAG,SAAS,CAAC;gBAClB,SAAS;YACV,CAAC;YACD,IAAI,SAAS,KAAK,GAAG,IAAI,CAAC,SAAS,KAAK,GAAG,IAAI,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;gBAC9D,OAAO,IAAI,CAAC;YACb,CAAC;YACD,SAAS;QACV,CAAC;QACD,IAAI,SAAS,KAAK,IAAI,EAAE,CAAC;YACxB,KAAK,EAAE,CAAC;YACR,SAAS;QACV,CAAC;QACD,IAAI,SAAS,KAAK,GAAG,EAAE,CAAC;YACvB,KAAK,GAAG,QAAQ,CAAC;YACjB,SAAS;QACV,CAAC;QACD,IAAI,SAAS,KAAK,GAAG,EAAE,CAAC;YACvB,KAAK,GAAG,QAAQ,CAAC;YACjB,SAAS;QACV,CAAC;QACD,IACC,SAAS,KAAK,IAAI;YAClB,SAAS,KAAK,IAAI;YAClB,SAAS,KAAK,GAAG;YACjB,SAAS,KAAK,GAAG;YACjB,SAAS,KAAK,GAAG;YACjB,SAAS,KAAK,GAAG;YACjB,SAAS,KAAK,GAAG;YACjB,SAAS,KAAK,GAAG;YACjB,CAAC,SAAS,KAAK,GAAG,IAAI,IAAI,KAAK,GAAG,CAAC,EAClC,CAAC;YACF,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IACD,OAAO,KAAK,KAAK,SAAS,CAAC;AAC5B,CAAC;AAED,MAAM,UAAU,wBAAwB,CACvC,OAAe;IAEf,MAAM,UAAU,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC;IAClC,IAAI,CAAC,UAAU,EAAE,CAAC;QACjB,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,gCAAgC,EAAE,CAAC;IAChE,CAAC;IACD,IACC,0BAA0B,CAAC,UAAU,CAAC;QACtC,4BAA4B,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAC7C,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,CACxB,EACA,CAAC;QACF,OAAO;YACN,EAAE,EAAE,KAAK;YACT,MAAM,EACL,gFAAgF;SACjF,CAAC;IACH,CAAC;IACD,IACC,CAAC,qBAAqB,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,EACjE,CAAC;QACF,OAAO;YACN,EAAE,EAAE,KAAK;YACT,MAAM,EACL,2FAA2F;SAC5F,CAAC;IACH,CAAC;IACD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;AACrB,CAAC;AAED,MAAM,UAAU,kBAAkB,CACjC,KAA6B;IAE7B,MAAM,QAAQ,GACb,KAAK,CAAC,MAAM,KAAK,mBAAmB;QACnC,CAAC,CAAC,+GAA+G;QACjH,CAAC,CAAC,+GAA+G,CAAC;IACpH,OAAO,iDAAiD,KAAK,CAAC,EAAE,OAAO,KAAK,CAAC,OAAO,WAAW,KAAK,CAAC,IAAI,yeAAye,QAAQ,EAAE,CAAC;AAC9lB,CAAC;AAED,MAAM,UAAU,wBAAwB,CACvC,KAA6B;IAE7B,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC;IAC5B,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,OAAO,cAAc,KAAK,CAAC,EAAE,KAAK,KAAK,CAAC,OAAO,KAAK,KAAK,CAAC,MAAM,aAAa,KAAK,CAAC,IAAI,EAAE,CAAC;IAC3F,CAAC;IACD,MAAM,IAAI,GAAG,CAAC,MAAgB,EAAE,KAAa,EAAE,EAAE,CAChD,MAAM,CAAC,MAAM;QACZ,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC;QAChD,CAAC,CAAC,KAAK,KAAK,EAAE,CAAC;IACjB,MAAM,QAAQ,GAAG,MAAM,CAAC,iBAAiB,CAAC;IAC1C,MAAM,gBAAgB,GAAG,QAAQ;QAChC,CAAC,CAAC,+CAA+C,QAAQ,CAAC,IAAI,YAAY,QAAQ,CAAC,GAAG,IAAI,iBAAiB,sBAAsB,QAAQ,CAAC,aAAa,EAAE,IAAI,CAAC,IAAI,CAAC,IAAI,MAAM,mBAAmB,QAAQ,CAAC,mBAAmB,EAAE,IAAI,CAAC,MAAM,CAAC,IAAI,MAAM,qBAAqB,QAAQ,CAAC,kBAAkB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,aAAa,wBAAwB,QAAQ,CAAC,eAAe,EAAE,IAAI,CAAC,IAAI,CAAC,IAAI,UAAU,2BAA2B,QAAQ,CAAC,kBAAkB,EAAE,IAAI,CAAC,IAAI,CAAC,IAAI,UAAU,yBAAyB,QAAQ,CAAC,gBAAgB,EAAE,IAAI,CAAC,IAAI,CAAC,IAAI,UAAU,gBAAgB,QAAQ,CAAC,aAAa,IAAI,SAAS,KAAK,QAAQ,CAAC,gBAAgB,IAAI,SAAS,kBAAkB,QAAQ,CAAC,cAAc,IAAI,SAAS,KAAK,QAAQ,CAAC,iBAAiB,IAAI,SAAS,kBAAkB,QAAQ,CAAC,cAAc,IAAI,SAAS,KAAK,QAAQ,CAAC,iBAAiB,IAAI,SAAS,GAAG;QAC30B,CAAC,CAAC,EAAE,CAAC;IACN,OAAO,yBAAyB,KAAK,CAAC,EAAE,KAAK,KAAK,CAAC,OAAO,iBAAiB,KAAK,CAAC,MAAM,uBAAuB,MAAM,CAAC,cAAc,qBAAqB,MAAM,CAAC,IAAI,sBAAsB,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE,sBAAsB,CAAC,iCAAiC,IAAI,CAAC,MAAM,CAAC,mBAAmB,EAAE,YAAY,CAAC,8BAA8B,IAAI,CAAC,MAAM,CAAC,gBAAgB,EAAE,eAAe,CAAC,qCAAqC,MAAM,CAAC,uBAAuB,GAAG,gBAAgB,EAAE,CAAC;AAC3d,CAAC"}
+{"version":3,"file":"assessment.js","sourceRoot":"","sources":["../src/assessment.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAe,MAAM,SAAS,CAAC;AAC5C,OAAO,EAAE,4BAA4B,EAAE,MAAM,aAAa,CAAC;AAE3D,MAAM,CAAC,MAAM,8BAA8B,GAC1C,6BAA6B,CAAC;AAC/B,MAAM,CAAC,MAAM,4BAA4B,GAAG,IAAI,CAAC,MAAM,CAAC;IACvD,IAAI,EAAE,IAAI,CAAC,MAAM,CAAC;QACjB,SAAS,EAAE,CAAC;QACZ,WAAW,EACV,0DAA0D;KAC3D,CAAC;CACF,CAAC,CAAC;AACH,MAAM,CAAC,MAAM,uCAAuC,GAAG,IAAI,CAAC,MAAM,CAAC;IAClE,IAAI,EAAE,IAAI,CAAC,MAAM,EAAE;IACnB,QAAQ,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;IACnC,mBAAmB,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;IAC9C,gBAAgB,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;IAC3C,uBAAuB,EAAE,IAAI,CAAC,MAAM,EAAE;IACtC,cAAc,EAAE,IAAI,CAAC,KAAK,CAAC;QAC1B,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC;QACvB,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC;QACtB,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC;KACtB,CAAC;IACF,iBAAiB,EAAE,IAAI,CAAC,QAAQ,CAAC,4BAA4B,CAAC;CAC9D,CAAC,CAAC;AA+BH,MAAM,UAAU,uBAAuB,CACtC,IAAY,EACZ,MAA+B,EAC/B,uBAAuB,GAAa,EAAE,EACtC,GAAG,GAAG,IAAI,IAAI,EAAE;IAEhB,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE;QAAE,MAAM,IAAI,KAAK,CAAC,6BAA6B,CAAC,CAAC;IACjE,OAAO;QACN,EAAE,EAAE,cAAc,GAAG,CAAC,OAAO,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,EAAE;QAC9C,OAAO,EAAE,CAAC;QACV,MAAM,EAAE,WAAW;QACnB,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE;QACjB,MAAM;QACN,UAAU,EAAE,GAAG,CAAC,WAAW,EAAE;QAC7B,uBAAuB;KACvB,CAAC;AACH,CAAC;AAED,MAAM,UAAU,wBAAwB,CACvC,KAA6B,EAC7B,MAA+B;IAE/B,OAAO;QACN,GAAG,KAAK;QACR,OAAO,EAAE,KAAK,CAAC,OAAO,GAAG,CAAC;QAC1B,MAAM,EAAE,UAAU;QAClB,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE;QACxB,MAAM;KACN,CAAC;AACH,CAAC;AAED,MAAM,UAAU,iBAAiB,CAChC,KAAyC;IAEzC,OAAO,KAAK,EAAE,MAAM,KAAK,WAAW,CAAC;AACtC,CAAC;AAED,MAAM,UAAU,kBAAkB,CACjC,KAA6B;IAE7B,OAAO,2CAA2C,KAAK,CAAC,IAAI,gVAAgV,CAAC;AAC9Y,CAAC;AAED,MAAM,UAAU,wBAAwB,CACvC,KAA6B;IAE7B,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC;IAC5B,IAAI,CAAC,MAAM;QAAE,OAAO,cAAc,KAAK,CAAC,EAAE,KAAK,KAAK,CAAC,IAAI,EAAE,CAAC;IAC5D,MAAM,IAAI,GAAG,CAAC,MAAgB,EAAE,EAAE,CACjC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC;QAC9C,iBAAiB,CAAC;IACnB,OAAO,2BAA2B,MAAM,CAAC,IAAI,uBAAuB,MAAM,CAAC,cAAc,sBAAsB,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,iCAAiC,IAAI,CAAC,MAAM,CAAC,mBAAmB,CAAC,8BAA8B,IAAI,CAAC,MAAM,CAAC,gBAAgB,CAAC,qCAAqC,MAAM,CAAC,uBAAuB,EAAE,CAAC;AACvU,CAAC"}

package/dist/enforcement/policy.d.ts

@@ -1,4 +1,5 @@
 import type { HarnessContract } from '../schema.js';
+export declare function pattern_matches(cwd: string, pattern: string, path: string): boolean;
 export declare function check_path_allowed(contract: HarnessContract, input_path: string): {
     ok: true;
 } | {
@@ -11,3 +12,4 @@ export declare function check_command_allowed(contract: HarnessContract, command
     ok: false;
     reason: string;
 };
+export declare function build_path_helpers(): string;