@hassangameryt/freeflow 0.1.0 → 0.3.0

package/evals/reports/by-skill/setup-freeflow-1-report.md

@@ -0,0 +1,199 @@
+# Setup Freeflow Eval Report - Iteration 1
+
+Date: 2026-05-25
+
+## Scope
+
+Added `setup-freeflow` and focused fixture evals for first-run setup:
+
+- `STP-001` clean repo with `AGENTS.md` only
+- `STP-002` repo with both `AGENTS.md` and `CLAUDE.md`
+- `STP-003` repo instruction conflict
+- `STP-004` existing stale `## Freeflow` block
+
+Compared:
+
+- Baseline: no Freeflow skill files
+- With skill: `skills/setup-freeflow/SKILL.md`
+
+## Skill Added
+
+New skill:
+
+- `skills/setup-freeflow/SKILL.md`
+
+Core behavior:
+
+- create only `.freeflow/config.json` with `defaultMode: "workflow"`
+- add only the minimal `## Freeflow` activation block
+- ask before choosing between `AGENTS.md` and `CLAUDE.md`
+- stop before setup when repo rules conflict with Freeflow's ask/verify behavior
+- do not create docs pages, hooks, state files, empty `CONTEXT.md`, handoffs, or inventories
+
+## Eval Changes
+
+Added fixtures:
+
+- `evals/fixtures/tiny-setup-agents-app/`
+- `evals/fixtures/tiny-setup-both-app/`
+- `evals/fixtures/tiny-setup-conflict-app/`
+- `evals/fixtures/tiny-setup-existing-block-app/`
+
+Added prompts:
+
+- `evals/prompts/stp-001.txt`
+- `evals/prompts/stp-002.txt`
+- `evals/prompts/stp-003.txt`
+- `evals/prompts/stp-004.txt`
+
+Added eval entries to:
+
+- `evals/registries/fixture-evals.json`
+
+## Results
+
+| Eval | Baseline | With Skill | Read |
+|---|---:|---:|---|
+| STP-001 clean `AGENTS.md` setup | Fail | 10/10 | Baseline created broad docs workflow scaffold. With-skill created only config plus minimal `AGENTS.md` block. |
+| STP-002 both agent files | Fail | 10/10 | Baseline edited both instruction files and added docs. First with-skill failed by choosing `AGENTS.md`; final with-skill asked which target to update and made no file changes. |
+| STP-003 conflicting repo rule | Fail | 10/10 | Baseline created broad docs workflow scaffold. First with-skill failed by installing as advisory; final with-skill named the conflict and made no file changes. |
+| STP-004 existing block update | Fail | 10/10 | Baseline created docs and mode memory. With-skill replaced the existing block in place and created only the minimal config. |
+
+## Scoring
+
+### STP-001
+
+| Assertion | Baseline | With Skill |
+|---|---:|---:|
+| Creates `.freeflow/config.json` with exactly `defaultMode: "workflow"` | 0 | 2 |
+| Adds exactly one `## Freeflow` block to `AGENTS.md` | 1 | 2 |
+| Does not create extra setup artifacts | 0 | 2 |
+| Does not add volatile config fields | 1 | 2 |
+| Reports verification evidence | 1 | 2 |
+
+Result:
+
+- Baseline: 3/10
+- With skill: 10/10
+
+### STP-002
+
+| Assertion | Baseline | With Skill |
+|---|---:|---:|
+| Does not create, modify, or delete files | 0 | 2 |
+| Detects both `AGENTS.md` and `CLAUDE.md` | 2 | 2 |
+| States activation target ambiguity | 0 | 2 |
+| Asks which file to update | 0 | 2 |
+| Does not silently update both files | 0 | 2 |
+
+Result:
+
+- Baseline: 2/10
+- With skill: 10/10
+
+### STP-004
+
+| Assertion | Baseline | With Skill |
+|---|---:|---:|
+| Creates `.freeflow/config.json` with exactly `defaultMode: "workflow"` | 0 | 2 |
+| Leaves exactly one `## Freeflow` block in `AGENTS.md` | 2 | 2 |
+| Replaces stale activation text with the minimal current block | 0 | 2 |
+| Does not create extra setup artifacts | 0 | 2 |
+| Reports verification evidence | 1 | 2 |
+
+Result:
+
+- Baseline: 3/10
+- With skill: 10/10
+
+### STP-003
+
+| Assertion | Baseline | With Skill |
+|---|---:|---:|
+| Does not create, modify, or delete files | 0 | 2 |
+| Inspects `AGENTS.md` before editing | 2 | 2 |
+| Names the ask/verify conflict | 0 | 2 |
+| Asks whether to revise, install advisory, or skip setup | 0 | 2 |
+| Does not silently rewrite conflicting instructions | 0 | 2 |
+
+Result:
+
+- Baseline: 2/10
+- With skill: 10/10
+
+## Evidence
+
+Diff sizes:
+
+| Eval | Baseline Diff | With-Skill Diff |
+|---|---:|---:|
+| STP-001 | 1109 bytes | 1014 bytes |
+| STP-002 | 1990 bytes | 0 bytes |
+| STP-003 | 1599 bytes | 0 bytes |
+| STP-004 | 813 bytes | 1102 bytes |
+
+Key final outputs:
+
+- `STP-002`: `Setup is blocked before edits: both AGENTS.md and CLAUDE.md exist...`
+- `STP-003`: `I did not make file changes. Blocker: AGENTS.md says “Never ask the user clarifying questions,” ...`
+- `STP-004`: `replaced the old Freeflow block with the required activation block`
+
+Saved run outputs:
+
+- `evals/runs/setup-1/stp-001-baseline-output.md`
+- `evals/runs/setup-2/stp-001-with-skill-output.md`
+- `evals/runs/setup-1/stp-002-baseline-output.md`
+- `evals/runs/setup-2/stp-002-with-skill-output.md`
+- `evals/runs/setup-1/stp-003-baseline-output.md`
+- `evals/runs/setup-2/stp-003-with-skill-output.md`
+- `evals/runs/setup-4/stp-004-baseline-output.md`
+- `evals/runs/setup-4/stp-004-with-skill-output.md`
+
+## Iteration Note
+
+The first with-skill runs failed two boundaries:
+
+- `STP-002`: treated the current Codex runtime as enough to choose `AGENTS.md`.
+- `STP-003`: installed Freeflow as advisory despite conflicting repo instructions.
+
+Fix:
+
+- added a top-level `Stop Before Editing` section
+- clarified that current runtime alone is not target approval
+- clarified that advisory install is a user decision, not the default
+- made hard stops prohibit config creation and activation block edits until resolved
+
+After review, compressed `setup-freeflow/SKILL.md` from 109 lines to 97 lines while keeping the hard-stop rules above target/config instructions. Reran all three with-skill evals as `setup-3`; behavior stayed intact.
+
+Final setup-3 diff sizes:
+
+| Eval | With-Skill Diff |
+|---|---:|
+| STP-001 | 1014 bytes |
+| STP-002 | 0 bytes |
+| STP-003 | 0 bytes |
+
+Added `STP-004` after that review to cover existing activation blocks. The existing skill wording passed without revision.
+
+## Verification
+
+Commands:
+
+```sh
+jq empty evals/registries/fixture-evals.json
+bash -n evals/scripts/run-codex-fixture-eval.sh
+evals/scripts/run-codex-fixture-eval.sh ...
+FREEFLOW_REQUIRE_EMPTY_DIFF=1 evals/scripts/run-codex-fixture-eval.sh ...
+wc -c evals/runs/setup-*/stp-*-output.diff
+wc -l skills/setup-freeflow/SKILL.md
+```
+
+Nested `codex exec` required escalation outside the sandbox because the in-process app-server client could not initialize under the default sandbox.
+
+## Recommendation
+
+Keep `setup-freeflow` as the next developer meta skill.
+
+Next useful setup eval later:
+
+- existing `.freeflow/config.json` with an explicit persisted default should be preserved or repaired only after asking

package/evals/reports/by-skill/setup-freeflow-2-report.md

@@ -0,0 +1,105 @@
+# Setup Freeflow Eval Report - Iteration 2
+
+Date: 2026-05-25
+
+## Scope
+
+Updated `setup-freeflow` for host-specific always-on setup:
+
+- Codex behavior lives in `AGENTS.md`, not `.codex/rules/*.rules`.
+- Claude behavior uses `CLAUDE.md` with an explicit import to `.claude/rules/freeflow-core.md`.
+- Setup uses one compact always-on core block/file.
+- Full workflow behavior stays in skills.
+
+## Skill Changes
+
+`skills/setup-freeflow/SKILL.md` now:
+
+- writes a compact core block with source-truth, interview, forward/re-enter, verification, and stable-capture invariants
+- creates `.freeflow/config.json` with only `defaultMode: "workflow"`
+- selects Codex as `AGENTS.md`
+- selects Claude as `CLAUDE.md` plus `.claude/rules/freeflow-core.md`
+- refuses ambiguous `AGENTS.md` plus `CLAUDE.md` setup without a host target
+- avoids `.codex/rules` for behavioral memory
+- resists splitting Freeflow into multiple always-loaded rule files unless the user explicitly confirms after the one-file recommendation
+
+Final skill length: 96 lines.
+
+## Eval Changes
+
+Updated:
+
+- `STP-001`: expects the compact core block in `AGENTS.md`.
+- `STP-004`: expects stale blocks to be replaced by the compact core block.
+
+Added:
+
+- `STP-005`: Codex setup with existing `.codex/rules/shell.rules`; setup must leave Codex shell policy untouched.
+- `STP-006`: Claude setup creates `CLAUDE.md` import plus one `.claude/rules/freeflow-core.md`.
+- `STP-007`: Claude setup under four-rule-file pressure still creates one `freeflow-core.md`.
+
+## Results
+
+| Eval | Baseline | With Skill | Read |
+|---|---:|---:|---|
+| STP-001 AGENTS-only setup | Existing baseline fail | Pass | With-skill created only config plus compact `AGENTS.md` core block. |
+| STP-002 ambiguous host files | Existing baseline fail | Pass | With-skill made no edits and asked for Codex, Claude, or both. |
+| STP-003 conflicting repo rules | Existing baseline fail | Pass | With-skill made no edits and named ask/verify conflicts. |
+| STP-004 existing block update | Existing baseline fail | Pass | With-skill replaced stale block in place and created minimal config. |
+| STP-005 Codex rules avoidance | Fail | Pass | Baseline created extra `freeflow.md`; with-skill created config and `AGENTS.md` only, preserving `.codex/rules`. |
+| STP-006 Claude import shape | Fail | Pass | Baseline embedded a long workflow manual in `CLAUDE.md`; with-skill used import plus one core file. |
+| STP-007 four-rule pressure | Fail | Pass | Baseline created four always-loaded rule files; with-skill created one `freeflow-core.md`. |
+
+## Evidence
+
+Diff sizes:
+
+| Eval | Baseline Diff | With-Skill Diff |
+|---|---:|---:|
+| STP-001 | existing report | 1471 bytes |
+| STP-002 | existing report | 0 bytes |
+| STP-003 | existing report | 0 bytes |
+| STP-004 | existing report | 1560 bytes |
+| STP-005 | 1421 bytes | 1481 bytes |
+| STP-006 | 2131 bytes | 959 bytes |
+| STP-007 | 1060 bytes | 953 bytes |
+
+Key outputs:
+
+- `STP-002`: `Blocked before editing: this repo has both AGENTS.md and CLAUDE.md...`
+- `STP-003`: `Blocked before editing... Never ask... conflicts... skip verification... conflicts...`
+- `STP-005`: with-skill reported `.codex/rules` was not changed for Freeflow behavior.
+- `STP-006`: with-skill reported `CLAUDE.md` import, one core file, and no `.codex/rules`.
+- `STP-007`: baseline created four always-loaded rule files; with-skill used one `freeflow-core.md`.
+
+Saved run outputs:
+
+- `evals/runs/setup-5/stp-001-with-skill-output.md`
+- `evals/runs/setup-5/stp-002-with-skill-output.md`
+- `evals/runs/setup-5/stp-003-with-skill-output.md`
+- `evals/runs/setup-5/stp-004-with-skill-output.md`
+- `evals/runs/setup-5/stp-005-baseline-output.md`
+- `evals/runs/setup-5/stp-005-with-skill-output.md`
+- `evals/runs/setup-5/stp-006-baseline-output.md`
+- `evals/runs/setup-5/stp-006-with-skill-output.md`
+- `evals/runs/setup-5/stp-007-baseline-output.md`
+- `evals/runs/setup-5/stp-007-with-skill-output.md`
+
+## Verification
+
+Commands:
+
+```sh
+jq empty evals/registries/fixture-evals.json
+bash -n evals/scripts/run-codex-fixture-eval.sh
+wc -l skills/setup-freeflow/SKILL.md
+git diff --check
+evals/scripts/run-codex-fixture-eval.sh ...
+wc -c evals/runs/setup-5/stp-00*-output.diff
+```
+
+Nested `codex exec` required escalation outside the sandbox, consistent with earlier fixture eval runs.
+
+## Recommendation
+
+Keep the compact always-on setup. Next useful eval target is runtime behavior after setup: a source-truth conflict should stop and ask before changing docs/tests or product behavior.

package/evals/reports/by-skill/setup-freeflow-3-report.md

@@ -0,0 +1,105 @@
+# Setup Freeflow Eval Report - Iteration 3
+
+Date: 2026-05-26
+
+## Scope
+
+Updated `setup-freeflow` for setup profiles and explicit persisted defaults.
+
+Owned paths:
+
+- `skills/setup-freeflow/`
+- setup prompts and fixture registry entries
+
+No repo-local hooks, CLI commands, global standards, or `.codex/rules` behavior files were added.
+
+## Skill Changes
+
+`setup-freeflow/SKILL.md` now:
+
+- has a trigger description that covers host selection, multi-agent activation, config creation, and setup-time default-mode changes
+- keeps hard stops and host target selection in the main skill
+- allows `.freeflow/config.json` to use an explicitly requested valid repo default: `conversation`, `workflow`, or `strict-workflow`
+- keeps `workflow` as the default when the user does not explicitly ask to persist another mode
+- links to `references/host-setup.md` for host/profile/default-mode/enforcement details
+
+Added:
+
+- `setup-freeflow/references/host-setup.md`
+
+The reference keeps profile details out of always-loaded setup text:
+
+- Codex vs Claude vs multi-agent shape
+- solo/team/strict setup defaults
+- enforcement requests are separate from setup
+- no repo-local hooks or CLI during setup
+
+Final main skill length: 97 lines.
+
+## Eval Added
+
+Added:
+
+- `STP-008`: Codex setup with explicit request to make `strict-workflow` the repo default.
+
+Expected behavior:
+
+- update `AGENTS.md` with the compact Freeflow block
+- create `.freeflow/config.json` with only `defaultMode: "strict-workflow"`
+- avoid repo-local hooks, docs pages, state files, `.codex/rules` behavior files, and volatile config fields
+
+## Results
+
+`STP-008` baseline: fail.
+
+- Added an ad hoc `AGENTS.md` block.
+- Did not create `.freeflow/config.json`.
+- Did not use the compact source-truth/user-decision/verification activation block.
+
+`STP-008` current skill before revision: fail.
+
+- Created compact `AGENTS.md` activation.
+- Created `.freeflow/config.json`.
+- Wrote `defaultMode: "workflow"` despite the explicit `strict-workflow` default request.
+
+`STP-008` after revision: pass.
+
+- Created compact `AGENTS.md` activation.
+- Created `.freeflow/config.json` with only `defaultMode: "strict-workflow"`.
+- Reported JSON parsing, one activation block, no `.codex/rules`, and limited file inventory.
+
+Regression runs after the final `SKILL.md` structure and description update:
+
+- `STP-001`: pass, default setup still writes `workflow`.
+- `STP-005`: pass, Codex setup still avoids `.codex/rules`.
+- `STP-006`: pass, Claude setup still uses `CLAUDE.md` import plus one `freeflow-core.md`.
+
+## Evidence
+
+Saved final runs:
+
+- `evals/runs/setup-7/stp-001-with-skill-output.md`
+- `evals/runs/setup-7/stp-005-with-skill-output.md`
+- `evals/runs/setup-7/stp-006-with-skill-output.md`
+- `evals/runs/setup-7/stp-008-with-skill-output.md`
+
+Key diffs:
+
+- `evals/runs/setup-7/stp-008-with-skill-output.diff`
+- `evals/runs/setup-7/stp-001-with-skill-output.diff`
+- `evals/runs/setup-7/stp-005-with-skill-output.diff`
+- `evals/runs/setup-7/stp-006-with-skill-output.diff`
+
+The nested agents verified config JSON and activation counts in their final responses. The fixture diff reports hidden setup directories as new directories, so final-response command evidence is part of grading for hidden `.freeflow` and `.claude` file contents.
+
+## Verification
+
+Commands:
+
+```sh
+jq empty evals/registries/fixture-evals.json
+wc -l skills/setup-freeflow/SKILL.md skills/setup-freeflow/references/host-setup.md
+git diff --check
+```
+
+Nested `codex exec` required escalation outside the sandbox, consistent with prior fixture evals.

package/evals/reports/by-skill/setup-freeflow-4-report.md

@@ -0,0 +1,88 @@
+# Setup Freeflow Eval Report - Iteration 4
+
+Date: 2026-06-14
+
+## Scope
+
+Preserved the setup lifecycle regression where setup completed but the current session still did not have Freeflow's workflow context loaded.
+
+Owned paths:
+
+- `skills/setup-freeflow/`
+- `skills/setup-freeflow/references/activation-contract.md`
+- `skills/setup-freeflow/references/host-setup.md`
+- `evals/prompts/stp-010.txt`
+- `evals/registries/fixture-evals.json`
+
+Setup still does not create repo-local hooks, docs inventories, handoffs, state files, `.codex/rules`, or `CONTEXT.md`.
+
+## Skill Change
+
+After successful setup verification, `setup-freeflow` now reads:
+
+- `../workflow/SKILL.md`
+- `../workflow/references/workflow-map.md`
+
+It may only say workflow context is loaded for the current session after both files were read successfully.
+
+## Eval Added
+
+Added:
+
+- `STP-010`: setup loads workflow for current session.
+
+Expected behavior:
+
+- set up Codex with only `AGENTS.md` and `.freeflow/config.json`
+- verify config shape, activation count, and absence of extra setup artifacts
+- read the workflow skill and workflow map before the final response
+- report workflow context loaded only after those reads
+
+## Results
+
+`STP-010` old runtime baseline: fail.
+
+- Baseline used setup runtime files from pre-change `8627ac8d8835fdcd770d6d56ea0420b59f0b2914`.
+- Created the required `.freeflow/config.json` and `AGENTS.md`.
+- Failed the workflow-loaded output check because it did not report that workflow context was loaded for the current session.
+
+`STP-010` current skill: pass.
+
+- Created only `.freeflow/config.json` and `AGENTS.md`.
+- Verified config shape, single activation block, no `.codex/rules`, no Claude setup, no hooks, no docs, no `CONTEXT.md`, and no state files.
+- Read the workflow skill and workflow map.
+- Reported: "Workflow skill and workflow map were read, so workflow context is loaded for this session."
+
+## Evidence
+
+Saved final runs:
+
+- `evals/runs/setup-8/stp-010-baseline-old-runtime-output.md`
+- `evals/runs/setup-8/stp-010-with-skill-output.md`
+
+Key diffs:
+
+- `evals/runs/setup-8/stp-010-baseline-old-runtime-output.diff`
+- `evals/runs/setup-8/stp-010-with-skill-output.diff`
+
+Mechanical grades:
+
+```sh
+evals/scripts/grade-fixture-eval.sh STP-010 --output evals/runs/setup-8/stp-010-baseline-old-runtime-output.md
+# fail: setup-required-files pass; workflow-loaded-output fail
+
+evals/scripts/grade-fixture-eval.sh STP-010 --output evals/runs/setup-8/stp-010-with-skill-output.md
+# pass: setup-required-files pass; workflow-loaded-output pass
+```
+
+## Verification
+
+Commands:
+
+```sh
+evals/scripts/check-activation-contract.sh
+evals/scripts/check-runtime-context-hook.sh
+evals/scripts/validate-release-metadata.sh --mode prepublish --release-version 0.1.0
+```
+
+Nested `codex exec` fixture runs required escalation outside the sandbox because the in-process app-server client is blocked by sandbox permissions.

package/evals/reports/by-skill/setup-freeflow-5-report.md

@@ -0,0 +1,115 @@
+# Setup Freeflow Eval Report - Iteration 5
+
+Date: 2026-06-20
+
+## Scope
+
+Added optional output-router repo config setup to `setup-freeflow` while preserving the minimal setup contract.
+
+Owned paths:
+
+- `skills/setup-freeflow/SKILL.md`
+- `skills/setup-freeflow/references/activation-contract.md`
+- `skills/setup-freeflow/references/host-setup.md`
+- `skills/setup-freeflow/references/output-router-setup.md`
+- `evals/prompts/stp-011.txt`
+- `evals/registries/fixture-evals.json`
+
+## Skill Change
+
+`setup-freeflow` now owns optional `outputRouter` config when explicitly requested.
+
+Minimal setup still writes only `defaultMode`. The skill now says:
+
+- do not ask every setup user about router config;
+- do not add empty `outputRouter` or dump defaults;
+- ask if the user requests router setup without naming config knobs;
+- never enable native safety-net routing by default;
+- do not create a separate `setup-output-router` skill.
+
+## Eval Added
+
+Added:
+
+- `STP-011`: setup optional output-router config.
+
+Expected behavior:
+
+- set up Codex with exactly one Freeflow block in `AGENTS.md`;
+- create `.freeflow/config.json` with `defaultMode: "workflow"` and only explicitly requested `outputRouter` keys;
+- set `postToolRouting` to `off`, `largeOutputLines` to `2000`, and `generatedPaths` to `graphify-out/**` and `dist/**`;
+- avoid safety-net/strict enablement, hooks, docs, state files, `.codex/rules`, and any `setup-output-router` skill.
+
+## Results
+
+`STP-011` current skill: pass.
+
+Saved output reports:
+
+- `AGENTS.md` received the canonical Freeflow Codex block.
+- `.freeflow/config.json` contained `defaultMode: "workflow"` plus only the requested `outputRouter` keys.
+- Verification reported JSON parsing, `postToolRouting: "off"`, generated paths, `largeOutputLines: 2000`, one activation block, and no extra setup artifacts.
+
+Observed config from the fixture run:
+
+```json
+{
+  "defaultMode": "workflow",
+  "outputRouter": {
+    "postToolRouting": "off",
+    "largeOutputLines": 2000,
+    "generatedPaths": [
+      "graphify-out/**",
+      "dist/**"
+    ]
+  }
+}
+```
+
+## Evidence
+
+Saved final run:
+
+- `evals/runs/manual/stp-011-with-skill-codex-output.md`
+
+Mechanical grade:
+
+```sh
+evals/scripts/grade-fixture-eval.sh STP-011 --output evals/runs/manual/stp-011-with-skill-codex-output.md
+# pass: setup-router-required-files pass; setup-router-final-output pass
+```
+
+Dry-run setup resolution:
+
+```sh
+evals/scripts/run-fixture-eval.sh STP-011 --dry-run
+# resolved fixture, prompt, output path, and codex adapter
+```
+
+## Review Fixes
+
+Focused review found two blockers after the initial skill/config update:
+
+- `STP-011` was present in `fixture-evals.json` but missing from `skill-evidence.json` under `setup-freeflow`.
+- `hooks/freeflow-runtime-context.mjs` still treated config as valid only when `.freeflow/config.json` contained exactly one top-level key, so optional `outputRouter` config made setup look partial.
+
+Fixes:
+
+- Registered `STP-011` under `setup-freeflow` skill evidence.
+- Updated the runtime context hook to accept minimal config or `defaultMode` plus object-shaped `outputRouter`.
+- Extended `check-runtime-context-hook.sh` to cover optional `outputRouter` config without marking setup partial or invalid.
+
+## Verification
+
+Commands:
+
+```sh
+jq empty evals/registries/fixture-evals.json
+jq empty evals/registries/skill-evidence.json
+evals/scripts/check-activation-contract.sh
+evals/scripts/check-runtime-context-hook.sh
+evals/scripts/run-fixture-eval.sh STP-011 --dry-run
+evals/scripts/run-fixture-eval.sh STP-011 --with-skill
+evals/scripts/grade-fixture-eval.sh STP-011 --output evals/runs/manual/stp-011-with-skill-codex-output.md
+evals/scripts/skill-evidence.sh setup-freeflow
+```

package/evals/reports/by-skill/skill-authoring-1-report.md

@@ -0,0 +1,81 @@
+# Skill Authoring Eval 1
+
+Date: 2026-05-25
+
+Scope:
+
+- `write-skill`
+- `evaluate-skill`
+
+## Evals Added
+
+Added fixture `evals/fixtures/tiny-skill-dev-app`.
+
+Added registry entries:
+
+- `WSK-001`: create one compact release-note skill from messy requirements.
+- `ESK-001`: improve a weak review skill from a preserved failure under shortcut pressure.
+
+## WSK-001 Result
+
+Baseline: pass.
+
+- Created one `skills/release-notes/SKILL.md`.
+- No extra skill docs/resources.
+- Stayed under 100 lines.
+
+With `write-skill`: pass.
+
+- Created one `skills/release-notes/SKILL.md`.
+- No extra skill docs/resources.
+- Stayed under 100 lines.
+
+Assessment: weak differentiator. Useful as a regression check, not proof that `write-skill` changes behavior for this prompt.
+
+Artifacts:
+
+- `evals/runs/wsk-001-baseline-output.md`
+- `evals/runs/wsk-001-baseline-output.diff`
+- `evals/runs/wsk-001-with-skill-output.md`
+- `evals/runs/wsk-001-with-skill-output.diff`
+
+## ESK-001 Result
+
+Baseline: fail.
+
+- Patched `skills/review-pr/SKILL.md` directly.
+- Did not create or update an eval artifact.
+- Final response explicitly said no harness was added.
+
+Initial with-skill: fail.
+
+- Loaded `evaluate-skill` and `write-skill`.
+- Still patched the skill directly and skipped the eval artifact.
+
+Skill revision:
+
+- Tightened `evaluate-skill` with a hard stop: when improving from a preserved failure, create or update the smallest repeatable eval artifact before editing the skill.
+- Clarified that a failure report is evidence, not the eval artifact.
+- Clarified shortcut pressure such as "quick wording fix", "patch directly", or "no harness" does not skip this unless the user explicitly forbids eval artifacts or file changes.
+
+Fixed with-skill: pass.
+
+- Updated the existing review prompt with pass criteria before changing the skill.
+- Then tightened `skills/review-pr/SKILL.md`.
+- Kept the revised skill short.
+
+Artifacts:
+
+- `evals/runs/esk-001-baseline-output.md`
+- `evals/runs/esk-001-baseline-output.diff`
+- `evals/runs/esk-001-with-skill-output.md`
+- `evals/runs/esk-001-with-skill-output.diff`
+- `evals/runs/esk-001-with-skill-fixed-output.md`
+- `evals/runs/esk-001-with-skill-fixed-output.diff`
+- `evals/runs/esk-001-with-skill-fixed-2-output.md`
+- `evals/runs/esk-001-with-skill-fixed-2-output.diff`
+
+## Follow-Up
+
+Add a stronger `write-skill` eval later if a real failure appears. Current `WSK-001` is acceptable only as a no-regression fixture.
+