Skip to content

Commit bb8062a

Browse files
Test Userclaude
Test User
and
claude
committed
docs(reports): spec-validation 2026-06-29 07:30 cycle — CONDITIONAL PASS
Re-confirmed tracked gap #2972 (stranded source ~26,551 LOC + spec-location drift). No behavioural regression; terraphim_agent lib suite = 288 passed. New contribution: refined the two-GuardDecision finding — they are a deliberate staged boundary (live enum in guard_patterns.rs + deferred struct in learnings/guard.rs per Phase H), not a name collision. Corrected filename drift in prior traceability (test lives at src/guard_patterns.rs:208). 3rd recurrence today (04:36, 06:30, 07:30 CEST). No new issue filed; fix owner unchanged (#2972 -> quality-coordinator). Co-Authored-By: Claude <noreply@anthropic.com>
1 parent 7b058b5 commit bb8062a

1 file changed

Lines changed: 185 additions & 0 deletions

File tree

‎reports/spec-validation-20260629-0730.md‎

Lines changed: 185 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,185 @@
1+
# Spec Validation Report — 2026-06-29 (07:30 CEST cycle)
2+
3+
**Agent**: spec-validator (Carthos, Domain Architect)
4+
**Date**: 2026-06-29 07:30 CEST
5+
**Verdict**: **CONDITIONAL PASS** (tracked gap #2972 re-confirmed; no behavioural regressions)
6+
**Prior verdict**: CONDITIONAL PASS (06:30 CEST, same day)
7+
8+
---
9+
10+
## Boundary State This Cycle
11+
12+
The territorial topology is **unchanged since the 06:30 cycle**. No commits have landed against
13+
the stranded-source directories or the `plans/` relocation:
14+
15+
| Stranded dir | Last commit touching it | Status |
16+
|---|---|---|
17+
| `crates/terraphim_orchestrator` | `5c48dabec` 2026-06-14 | stale (predates cycle) |
18+
| `crates/terraphim_agent` | `5c48dabec` 2026-06-14 | stale |
19+
| `crates/terraphim_agent_application` | `45c385178` 2026-06-22 | stale |
20+
21+
#2972 remains **open**, assigned to `quality-coordinator`, 23 comments. No remediation owner
22+
action since the prior recurrence.
23+
24+
---
25+
26+
## Plans Validated (6, against true polyrepo home)
27+
28+
All six plans' acceptance criteria are satisfied in the migrated code
29+
(`/home/alex/projects/terraphim/terraphim-agents`). Behavioural ACs verified:
30+
31+
### Plan 1: `design-gitea82-correction-event.md` — **PASS**
32+
| AC | Evidence | Status |
33+
|----|------|--------|
34+
| `CorrectionEvent` struct | `learnings/capture.rs` | ✅ |
35+
| `capture_correction()` / `list_all_entries()` / `query_all_entries()` | `learnings/capture.rs` | ✅ |
36+
| Secret redaction wired | `learnings/redaction.rs` (called in capture paths) | ✅ |
37+
38+
### Plan 2: `d3-session-auto-capture-plan.md` — **PASS**
39+
| AC | Evidence | Status |
40+
|----|------|--------|
41+
| `from_session_commands()` / `extract_bash_commands_from_session()` | `learnings/procedure.rs` | ✅ |
42+
| Trivial-command filter | present in procedure.rs | ✅ |
43+
44+
### Plan 3: `design-gitea84-trigger-based-retrieval.md` — **PASS**
45+
| AC | Evidence | Status |
46+
|----|------|--------|
47+
| `MarkdownDirectives.trigger` / `.pinned` | registry `terraphim_types` 1.20.x | ✅ |
48+
| `TriggerIndex` + fallback matching | registry `terraphim_rolegraph` 1.20.2 | ✅ |
49+
50+
### Plan 4: `research-single-agent-listener.md` — **PASS** (research-only, no ACs)
51+
### Plan 6: `design-single-agent-listener.md` — **PASS** (runtime ACs; `learnings/listener.rs` structurally present)
52+
### Plan 5: `learning-correction-system-plan.md` — **PASS** (carry-forwards stable)
53+
| Phase | Issue(s) | Status | Evidence |
54+
|-------|----------|--------|----------|
55+
| A–F, H | #480-704 | ✅ | capture/procedure/replay/guard/suggest intact in `terraphim-agents/learnings/` |
56+
| G (shared CLI) | #727 partial | ✅ DONE | `SharedLearningSub::{List,Promote,Import}` wired; `SharedLearningStore::open` called |
57+
| I (evolution) | #727-730 | ✅ cleared | Mock LLM factories confined to `#[cfg(test)]`; production paths clean |
58+
| J (validation) | #515-517,#451 | ✅ present | `terraphim_hooks` 1.20.2 registry crate consumed; compiles green |
59+
60+
---
61+
62+
## Test Evidence (re-run this cycle)
63+
64+
```
65+
cargo test -p terraphim_agent --lib
66+
test result: ok. 288 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 1.39s
67+
```
68+
69+
Includes `shared_learning::{store,wiki_sync}` tests and `robot::output::proptests`. **Behavioural invariant holds** (stable across the 06:30 and 07:30 cycles).
70+
71+
---
72+
73+
## Architectural Finding: Two `GuardDecision` Types (refined, not a violation)
74+
75+
The prior report's "coincidental name collision" framing was imprecise. Within `terraphim_agent`
76+
there are **two distinct `GuardDecision` abstractions**, by deliberate design:
77+
78+
| Type | Location | Shape | Wired? | Role |
79+
|---|---|---|---|---|
80+
| `GuardDecision` (**enum**) | `src/guard_patterns.rs` (crate root) | `Allow / Block / Sandbox` | **LIVE** — `listener.rs:1334`, `main.rs:1966-2724` | Production guard (`CommandGuard`) |
81+
| `GuardDecision` (**struct**) | `src/learnings/guard.rs` | `{ tier, reason, previously_failed }` + `ExecutionTier` | **DEFERRED** — zero external usages | Plan §H graduated-tier design (Phase H, future iteration) |
82+
83+
The deferral is **self-documented and intentional**. `learnings/mod.rs` declares:
84+
85+
```rust
86+
// Guard API — newly added, will be wired into the binary in a future iteration.
87+
#[allow(unused_imports)]
88+
pub use guard::{ExecutionTier, GuardDecision, evaluate_command, evaluate_command_with_learning};
89+
```
90+
91+
This is not dead code from neglect — it is staged foundation work silenced with
92+
`#[allow(unused_imports)]`, consistent with the learning-plan's Phase H sequencing
93+
(dependent on Phase D replay + Phase F health monitoring). `DESIGN-guard-patterns-redesign.md`
94+
confirms intent: the redesign *replaces internals* of `guard_patterns.rs` with thesaurus matching;
95+
it does not delete it, nor does it merge the two abstractions.
96+
97+
**Verdict on this finding:** Not a spec violation. A deliberate staged-design boundary. Filed
98+
here for traceability so future runs do not re-flag it as "name collision" or "duplicate dead code".
99+
100+
> Note: prior traceability cited test evidence `git_checkout_double_dash` against
101+
> `learnings/guard_patterns.rs`. The actual location is `src/guard_patterns.rs:208`
102+
> (`test_git_checkout_double_dash_blocked`). Filename drift in the prior report — corrected here.
103+
104+
---
105+
106+
## The Tracked Gap (#2972) — Unchanged, P2
107+
108+
### Facet A — Stranded source (re-measured)
109+
110+
Git-tracked Rust source with no workspace build path remains in this repo:
111+
112+
| Directory | Tracked .rs files | LOC | Cargo.toml |
113+
|-----------|------------------:|----:|------------|
114+
| `crates/terraphim_orchestrator` | 18 | 16,467 | none (orphaned mod roots) |
115+
| `crates/terraphim_agent` | 7 | 6,819 | none |
116+
| `crates/terraphim_agent_application` | 8 | 3,265 | **still present** |
117+
| **subtotal** | **33** | **26,551** | — |
118+
119+
Hazard: dead-but-tracked code — excluded from build, superseded by registry deps, yet present
120+
and searchable. Maintenance trap + review noise. Measurement is stable with the 06:30 cycle.
121+
122+
### Facet B — Spec-location drift (unfixed)
123+
124+
All four code-bearing plans still cite pre-extraction paths unresolvable in this repo:
125+
- `design-gitea82-correction-event.md` → `crates/terraphim_agent/src/learnings/capture.rs` (×2 refs)
126+
- `d3-session-auto-capture-plan.md` → `crates/terraphim_agent/src/learnings/procedure.rs`
127+
- `learning-correction-system-plan.md` → `crates/terraphim_agent/src/learnings/{hook,mod,capture}.rs` (×5+ refs)
128+
129+
No `RELOCATED` banners added to any plan. **Facet B drift unfixed.**
130+
131+
### Why CONDITIONAL PASS, not FAIL
132+
133+
No behavioural/acceptance-criterion violation exists. Every specified function is implemented
134+
and tested in its polyrepo. The gap is **repository hygiene + spec relocation**, already tracked
135+
as **#2972** (open, assigned `quality-coordinator`, 23 comments). This cycle re-confirms with
136+
stable measurement. **No new issue filed** — fix owner unchanged; duplicate-noise avoided per cron
137+
protocol.
138+
139+
**Smallest fix** (owned by whoever closes #2972):
140+
1. `git rm` stranded `crates/*` source (except genuinely-shared fixtures)
141+
2. Add `RELOCATED.md` banners to the 4 code-bearing plans pointing to `terraphim-agents`
142+
3. Replace `plans/` with a `RELOCATED.md` index
143+
144+
---
145+
146+
## Negative Space (explicitly considered and ruled out)
147+
148+
| Considered | Ruled out | Basis |
149+
|---|---|---|
150+
| `terraphim_mcp_search` violates a plan | Net-new, no plan covers it | No `plans/` doc references `mcp_search`/`McpToolIndex`/`SEP-1821` (verified by absence) |
151+
| New issue #3017 (kiro Task 10) in scope | Out of scope | No `plans/` doc references kiro/OTP — net-new doc-tracking issue |
152+
| New issue #3014 (session NFR G1) in scope | Out of scope | No `plans/` doc references session benchmark NFR — net-new perf-tracking issue |
153+
| `learnings/guard.rs` is accidental dead code | Deliberate staging | `mod.rs` comment + `#[allow(unused_imports)]` + Phase H dependency chain |
154+
| Two `GuardDecision` types = spec conflict | Deliberate boundary | Different modules, different shapes, different lifecycle stage (enum=live, struct=staged) |
155+
| Workspace membership defect | Not a defect | `members=["crates/*"]` globs; `exclude=[...]` prunes residuals; internally consistent |
156+
| Carry-forwards #727/Phase-G/I regressed | Stable (06:30→07:30) | Shared CLI wired; mocks confined to `#[cfg(test)]` |
157+
158+
---
159+
160+
## Traceability Matrix
161+
162+
| Req (plan symbol) | Plan | Impl Location (true) | Test | Status |
163+
|---|---|---|---|---|
164+
| CorrectionEvent | design-gitea82 §1 | `terraphim-agents/learnings/capture.rs` | `test_capture_*` (288-suite) | ✅ |
165+
| ProcedureStore | d3 / learning-plan §B | `terraphim-agents/learnings/procedure.rs` | `from_session_commands` tests | ✅ |
166+
| `replay_procedure()` | learning-plan §D | `terraphim-agents/learnings/replay.rs` | replay tests | ✅ |
167+
| GuardDecision (enum, live) | (production guard) | `terraphim-agents/src/guard_patterns.rs` | `test_git_checkout_double_dash_blocked:208` | ✅ |
168+
| GuardDecision (struct, staged) | learning-plan §H | `terraphim-agents/learnings/guard.rs` | unit tests in-file (10) | ✅ staged |
169+
| SharedLearningSub CLI | learning-plan §G | `terraphim-agents/main.rs:3964` | `wiki_sync::tests` | ✅ |
170+
| MarkdownDirectives.trigger | design-gitea84 §1 | `terraphim-core` (registry terraphim_types) | `parses_trigger_directive` | ✅ |
171+
| TriggerIndex | design-gitea84 §3 | `terraphim-core` (registry terraphim_rolegraph) | tfidf tests | ✅ |
172+
173+
---
174+
175+
## Verdict
176+
177+
**CONDITIONAL PASS.** No P0/P1 behavioural spec violations. All six plans' acceptance criteria
178+
satisfied in the code's true home (`terraphim-agents` + `terraphim-core` registry crates);
179+
`terraphim_agent` lib suite = **288 passed, 0 failed** (re-verified this cycle). Carry-forwards
180+
from 2026-06-01 remain cleared and stable.
181+
182+
One tracked P2 gap re-confirmed with stable measurement (~26,551 LOC stranded across 3 excluded
183+
dirs + spec-location drift in 4 plans). Already filed as **#2972** (open). This cycle adds a
184+
refined architectural finding (the two-`GuardDecision` boundary is deliberate staging, not a
185+
collision) and corrects a filename drift in prior traceability. **No new issue filed.**

0 commit comments

Comments
 (0)