feat(roles): implement F100 dedicated roles/ namespace for agent role discovery

.gitignore

@@ -57,3 +57,7 @@ site/node_modules/
 examples/plugins/*/awf-plugin-*
 !examples/plugins/*/awf-plugin-*.go
 !examples/plugins/*/awf-plugin-*_test.go
+
+# Added by awf project-onboard
+.awf/storage/states/
+.awf/storage/logs/

.zpm/kb/default/knowledge.pl

@@ -0,0 +1,304 @@
+% base-schema.pl — Shared AWF knowledge-base schema.
+% Loaded automatically from .zpm/kb/default/.
+
+% Dynamic declarations — required by Trealla Prolog so that rules referencing
+% these predicates compile even before any facts are asserted at runtime.
+:- dynamic(task/3).
+:- dynamic(depends_on/2).
+:- dynamic(task_complete/1).
+:- dynamic(task_file/2).
+:- dynamic(component/4).
+:- dynamic(stub/3).
+:- dynamic(test_file/2).
+:- dynamic(source_file/1).
+:- dynamic(adr/3).
+:- dynamic(workflow_run/3).
+:- dynamic(impl_note/3).
+:- dynamic(feedback_rule/3).
+:- dynamic(imports/2).
+:- dynamic(layer/2).
+:- dynamic(uses_panic/1).
+:- dynamic(coverage/2).
+:- dynamic(integrity_violation/2).
+
+% --- Task graph (written by plan, read by implement/fix-errors) ---
+% task(Id, Title, Status).  Status ∈ {pending, in_progress, completed, blocked}.
+% depends_on(TaskA, TaskB).  TaskA depends on TaskB completing first.
+% task_complete(Id).          Written by implement when stubs cleared + tests green.
+% incomplete_task(Id).        Derived: task that is pending/in_progress with unresolved stubs.
+
+incomplete_task(Id) :-
+    task(Id, _, Status),
+    Status \= completed,
+    stub(File, _, critical),
+    task_file(Id, File).
+
+% --- Component graph (written by plan, read by implement) ---
+% component(Id, Name, Path, Type).  Type ∈ {module, file, package, pipeline}.
+% pipeline_handled(Path).           Paths managed by pipelines (CHANGELOG, docs/...).
+% forbidden_component(Id) :-        Components plan must not include.
+%     component(Id, _, Path, _), pipeline_handled(Path).
+
+% forbidden_component/1: Path is bound via component/4, so nonvar-guarded
+% pipeline_handled/1 prefix rules fire correctly. Do not call
+% pipeline_handled(X) with X unbound — rules will not enumerate.
+forbidden_component(Id) :-
+    component(Id, _, Path, _),
+    pipeline_handled(Path).
+
+% Seed the standard pipeline-handled paths (exact matches)
+pipeline_handled('CHANGELOG.md').
+pipeline_handled('docs/CHANGELOG.md').
+pipeline_handled('README.md').
+pipeline_handled('README').
+pipeline_handled('CHANGELOG').
+pipeline_handled('CLAUDE.md').
+
+% Prefix-based rules (match directories handled by pipelines).
+% nonvar/1 guard prevents ZPM 0.1.0 instantiation errors when Path is unbound,
+% which would otherwise abort enumeration of the entire predicate (see KNOWN-ISSUES).
+% These rules only fire when Path is already bound (the normal use case via forbidden_component/1).
+pipeline_handled(Path) :- nonvar(Path), sub_atom(Path, 0, _, _, 'docs/').
+pipeline_handled(Path) :- nonvar(Path), sub_atom(Path, 0, _, _, '.specify/memory/').
+pipeline_handled(Path) :- nonvar(Path), sub_atom(Path, 0, _, _, 'ADR/').
+
+% --- Stub tracking (written by code, read by implement) ---
+% stub(File, Line, Kind).  Kind ∈ {critical, cosmetic, todo, fixme}.
+% task_file(TaskId, File).  Links stubs back to tasks.
+
+% --- Test coverage (written by plan/implement, used by audit) ---
+% test_file(SourcePath, TestPath).
+% covered_by(Source, Test) :- test_file(Source, Test).
+
+% Decision: wrap test_file/2 in catch/3, same reason as current_decision.
+% Why: ZPM (Scryer Prolog) raises existence_error when a constitution rule
+% evaluates \+ covered_by(File, _) on a fresh KB with zero test_file facts.
+% catch/3 intercepts that error and cleanly fails, so the negation succeeds
+% ("no test covers this source") and missing_test violations derive correctly.
+covered_by(Source, Test) :-
+    catch(test_file(Source, Test), _, fail).
+
+% --- ADR (written by plan/spec-generate, read by architecture reviews) ---
+% adr(Id, Title, Decision).
+% supersedes(NewId, OldId).
+% current_decision(Decision) :-
+%     adr(Id, _, Decision), \+ supersedes(_, Id).
+
+% Decision: use catch/3 instead of \+/1 for supersedes check.
+% Why: ZPM (Scryer Prolog) raises existence_error when \+ calls an
+% undefined predicate. catch/3 handles both the undefined-predicate case
+% (no supersedes facts yet → succeed) and the normal case (supersedes
+% facts exist → \+ evaluates correctly).
+% Trade-off: slightly less idiomatic Prolog; necessary for ZPM compatibility.
+current_decision(Id, Decision) :-
+    adr(Id, _, Decision),
+    catch((\+ supersedes(_, Id)), _, true).
+
+% --- Feedback rules (written by feedback workflow, read by review agents) ---
+% feedback_rule(Topic, Date, Text).
+%   Topic: atom matching CLAUDE.md section slug (e.g. architecture_rules)
+%          derived from section name: lowercase, spaces→underscores
+%   Date:  ISO 8601 atom 'YYYY-MM-DD' — lexicographic comparison works for
+%          date ordering, e.g. Date @>= '2026-04-15' for "last 7 days"
+%   Text:  quoted atom with the rule body
+% Query "recent feedback" example (caller computes cutoff date):
+%   feedback_rule(Topic, Date, Text), Date @>= '2026-04-15'.
+
+% --- Workflow history (written by commit) ---
+% workflow_run(Name, Date, Status).  Status ∈ {success, failure, partial}.
+
+% --- LLM-emitted rationale (written by plan/code-review agents, read by audit) ---
+% Captured from fenced ```zpm blocks emitted by agents; see
+% scripts/zpm/parse-zpm-block.sh.
+%
+% task_rationale(TaskId, Text).
+%   Why this task exists — captured at generate_tasks time.
+%   Keyed by TaskId (use upsert: one rationale per task).
+%
+% task_risk(TaskId, Level, Reason).
+%   Level ∈ {low, medium, high}.
+%   LLM's self-assessed risk. Keyed by TaskId (use upsert).
+%
+% review_finding(Issue, File, Line, Severity, Kind).
+%   Severity ∈ {info, warning, critical}.
+%   Kind: short atom (e.g. 'duplication', 'n_plus_1', 'hard_coded_secret').
+%   Multi-valued — many findings per issue. Use assert.
+%
+% impl_note(File, Symbol, Note).
+%   Local implementation decisions captured during the code TDD cycle.
+%   Multi-valued — many notes per (File, Symbol). Use assert.
+
+% Convenience rule: high-risk pending tasks (for audit and dashboards).
+high_risk_open_task(Id, Title) :-
+    task(Id, Title, Status),
+    Status \= completed,
+    task_risk(Id, high, _).
+
+% --- Generic integrity violation scaffolding ---
+% Constitutions assert integrity_violation/2 rules.
+% Severity ∈ {error, warning, info}.
+% Example (from constitution-go.pl):
+%   integrity_violation(layer_inversion, File) :-
+%     imports(File, Target),
+%     layer(File, domain),
+%     layer(Target, infrastructure).
+
+% ─── Project Memory ──────────────────────────────────────────────────────────
+:- dynamic(observation/4).
+:- dynamic(decision/4).
+:- dynamic(convention/3).
+
+% observation(Id, Category, Content, Date)
+%   Id: unique atom (auto-generated slug)
+%   Category: pattern | convention | quirk | dependency | performance
+%   Content: description (single-quoted atom)
+%   Date: ISO date atom e.g. '2026-05-24'
+%
+% decision(Id, What, Why, TradeOff)
+%   Architectural/design decisions recorded during workflows
+%
+% convention(Id, Pattern, Context)
+%   Discovered codebase conventions worth preserving
+
+observations_by_category(Cat, Ids) :-
+    findall(Id, observation(Id, Cat, _, _), Ids).
+
+decisions_about(Topic, Ids) :-
+    findall(Id, (decision(Id, What, _, _), sub_atom(What, _, _, _, Topic)), Ids).
+% constitution-go.pl — Go project integrity rules (Prolog).
+% Loaded automatically from .zpm/kb/ alongside base-schema.pl.
+%
+% Encodes a machine-checkable subset of Project Constitution v1.0.0 (go.md).
+% Rules produce integrity_violation(Kind, File) facts that audit tooling queries.
+%
+% ---  7 Principles (verbatim titles + 1-line summaries from go.md)  ----------
+%
+% P1 — Hexagonal Architecture
+%      Domain depends on nothing; application on domain; infrastructure
+%      implements ports; interfaces depend on application.
+%
+% P2 — Go Idioms
+%      No panic in library code; interfaces defined by consumer; accept
+%      interfaces return structs; explicit error handling.
+%
+% P3 — Test-Driven Development
+%      RED → GREEN → REFACTOR; 80% minimum code coverage; table-driven tests.
+%
+% P4 — Error Taxonomy
+%      Exit codes 1-4 map to user / config / execution / system errors.
+%      (Not modelled: purely a runtime/CLI convention, not statically checkable)
+%
+% P5 — Security First
+%      Inputs untrusted; secrets masked; atomic writes; no plaintext creds.
+%      (Not modelled: requires data-flow analysis, beyond static ZPM facts)
+%
+% P6 — Minimal Abstraction
+%      No premature optimisation; delete unused code; comments only when needed.
+%      (Not modelled: heuristic too weak without AST; no reliable Prolog rule)
+%
+% P7 — Documentation Co-location
+%      README.md, CLAUDE.md, CHANGELOG.md, specs in .specify/.
+%      (Not modelled: file-presence checks handled by shell audit scripts)
+%
+% ---------------------------------------------------------------------------
+%
+% FACT SCHEMA (asserted by callers, e.g. via zpm-assert.sh):
+%
+%   layer(File, Layer)          File belongs to Layer.
+%                               Layer ∈ {domain, application, infrastructure, interfaces}
+%   imports(File, Target)       File depends on Target (path or module import).
+%   source_file(File)           File is a Go source file tracked in the project.
+%   coverage(File, Pct)         File has measured test coverage Pct (integer 0-100).
+%   uses_panic(File)            File contains a panic() call.
+%
+%   covered_by/2 is defined in base-schema.pl via test_file/2 — do NOT redefine here.
+%
+% ---------------------------------------------------------------------------
+
+% layer_index/2: numeric order of layers, inner (low) → outer (high).
+% Used to detect when a lower-indexed layer imports a higher-indexed layer.
+
+layer_index(domain,         0).
+layer_index(application,    1).
+layer_index(infrastructure, 2).
+layer_index(interfaces,     3).
+
+% --- Violation kinds modelled in this constitution -------------------------
+% Enumeration helper for audit tooling. Without this, querying
+% integrity_violation(K, F) with both vars unbound returns [] in ZPM 0.1.0.
+violation_kind(layer_inversion).
+violation_kind(missing_test).
+violation_kind(framework_in_domain).
+violation_kind(panic_in_library).
+violation_kind(low_coverage).
+
+% --- P1: Layer inversion ----------------------------------------------------
+% Violation when a file in an inner layer imports a file in an outer layer.
+
+integrity_violation(layer_inversion, File) :-
+    imports(File, Target),
+    layer(File, LayerA),
+    layer(Target, LayerB),
+    layer_index(LayerA, IdxA),
+    layer_index(LayerB, IdxB),
+    IdxA < IdxB.
+
+% --- P3: Missing test -------------------------------------------------------
+% A source_file has no covered_by entry (via base-schema.pl test_file/2).
+% Excludes main.go (entry point) and _test.go files (they ARE the tests).
+
+is_main_file('main.go').
+is_main_file(File) :- atom_concat(_, '/main.go', File).
+is_main_file(File) :- atom_concat(_, '_test.go', File).
+
+integrity_violation(missing_test, File) :-
+    source_file(File),
+    \+ covered_by(File, _),
+    \+ is_main_file(File).
+
+% --- P2: Framework import in domain layer -----------------------------------
+% Domain code must not import web / ORM frameworks (tight coupling to infra).
+
+web_framework('github.com/gin-gonic/gin').
+web_framework('github.com/labstack/echo').
+web_framework('github.com/labstack/echo/v4').
+web_framework('gorm.io/gorm').
+web_framework('github.com/gofiber/fiber/v2').
+web_framework('github.com/go-chi/chi/v5').
+web_framework('github.com/gorilla/mux').
+
+integrity_violation(framework_in_domain, File) :-
+    layer(File, domain),
+    imports(File, Target),
+    web_framework(Target).
+
+% --- P2: Panic in library code ----------------------------------------------
+% Library layers (domain, application, infrastructure) must not use panic.
+% The interfaces layer (CLI/API entry points) is exempt — it may panic on
+% unrecoverable startup errors.
+
+library_layer(domain).
+library_layer(application).
+library_layer(infrastructure).
+
+integrity_violation(panic_in_library, File) :-
+    uses_panic(File),
+    layer(File, Layer),
+    library_layer(Layer).
+
+% --- P3: Low coverage -------------------------------------------------------
+% Any file with a coverage fact below 80% is a violation.
+
+integrity_violation(low_coverage, File) :-
+    coverage(File, Pct),
+    Pct < 80.
+
+% --- Skipped rules ----------------------------------------------------------
+%
+% P5 — logic_in_interfaces (draft rule from plan §563-630):
+%   The heuristic "interface file has >N lines of non-declaration code" is
+%   unreliable without a proper AST. Skipped; catches too many false positives.
+%
+% P5 — package_name_mismatch:
+%   Go package naming is enforced by `go vet` and the compiler. Reproducing
+%   that logic in ZPM provides no additional safety signal.

.zpm/kb/feedback/knowledge.pl

@@ -0,0 +1,62 @@
+:- module(feedback, []).
+% ─── Feedback Rules Schema ───────────────────────────────────────────────────
+% Memory segment: feedback
+% Lifecycle: created at project setup, accumulates across workflow runs.
+% Replaces markdown-based feedback in CLAUDE.md with queryable Prolog facts.
+%
+% Facts:
+%   rule(Id, Category, Description, Priority, Source)
+%     Id: unique atom (e.g. rule_001, rule_avoid_sql_concat)
+%     Category: architecture | pitfall | test | review | style
+%     Description: the rule text (single-quoted atom)
+%     Priority: high | medium | low
+%     Source: implement | audit | fix_errors | pr_review
+%
+%   example(RuleId, Type, Code, Explanation)
+%     Type: good | bad
+%
+%   trigger(RuleId, Pattern, Scope)
+%     Pattern: file/path substring to match
+%     Scope: file | directory | project
+%
+% Dynamic declarations (required by Trealla Prolog for runtime assertion).
+:- dynamic(rule/5).
+:- dynamic(example/4).
+:- dynamic(trigger/3).
+
+% Derived rules:
+
+% A rule is applicable to a file if its trigger pattern matches.
+applicable(RuleId, File) :-
+    trigger(RuleId, Pattern, file),
+    sub_atom(File, _, _, _, Pattern).
+applicable(RuleId, File) :-
+    trigger(RuleId, Pattern, directory),
+    sub_atom(File, _, _, _, Pattern).
+applicable(RuleId, _File) :-
+    trigger(RuleId, _, project).
+
+% Find all rules effective for a list of files (deduplicated).
+effective_rules(Files, RuleIds) :-
+    findall(Id, (member(F, Files), applicable(Id, F)), All),
+    sort(All, RuleIds).
+
+% Get high-priority rules.
+priority_rules(Ids) :-
+    findall(Id, rule(Id, _, _, high, _), Ids).
+
+% Rules by category.
+rules_by_category(Cat, Ids) :-
+    findall(Id, rule(Id, Cat, _, _, _), Ids).
+
+% Rules by source workflow.
+rules_by_source(Src, Ids) :-
+    findall(Id, rule(Id, _, _, _, Src), Ids).
+
+% Full rule detail for display.
+rule_detail(Id, Cat, Desc, Prio) :-
+    rule(Id, Cat, Desc, Prio, _).
+
+% Count rules.
+rule_count(N) :-
+    findall(Id, rule(Id, _, _, _, _), L), length(L, N).

.zpm/mounts.json

@@ -6,6 +6,12 @@
       "path": ".zpm/kb/default",
       "scope": "project",
       "mode": "rw"
+    },
+    {
+      "name": "feedback",
+      "path": ".zpm/kb/feedback",
+      "scope": "project",
+      "mode": "rw"
     }
   ]
 }