@ax-skills/all 0.1.0 → 0.1.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/bin/install.mjs
CHANGED
|
@@ -53,7 +53,7 @@ function runInstaller() {
|
|
|
53
53
|
console.log(`[@ax-skills/all] Installed ${entry.name} -> ${path.relative(process.cwd(), destDir)}`);
|
|
54
54
|
}
|
|
55
55
|
|
|
56
|
-
console.log(`[@ax-skills/all@0.1.
|
|
56
|
+
console.log(`[@ax-skills/all@0.1.1] Installed ${entries.length} skill(s) into ${targetRelDir}`);
|
|
57
57
|
}
|
|
58
58
|
|
|
59
59
|
runInstaller();
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@ax-skills/all",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.1",
|
|
4
4
|
"description": "Complete Agent Experience (AX) skill suite combining Software Factory workflow skills, agent-ergonomic software engineering, and React architecture standards.",
|
|
5
5
|
"license": "(MIT AND CC-BY-4.0)",
|
|
6
6
|
"author": "Aaron Peloquin (https://agent-experience.ai)",
|
|
@@ -7,7 +7,7 @@ description: Process document for conducting an independent pre-human proof pass
|
|
|
7
7
|
|
|
8
8
|
## Role & Editorial Proofing Framing
|
|
9
9
|
|
|
10
|
-
Act as an Independent Proofing Editor and Adversarial Red-Team Lead inspecting a completed body of work in a fresh context window (or as an isolated sub-agent) right before human review. Apply three complementary lenses in a single constructive pass: **proofreading** qualitative craft against domain standards, verifying **proof as evidence** against the original requirements and completion notes, and compiling a **pass** readiness briefing that catches
|
|
10
|
+
Act as an Independent Proofing Editor and Adversarial Red-Team Lead inspecting a completed body of work in a fresh context window (or as an isolated sub-agent) right before human review. Apply three complementary lenses in a single constructive pass: **proofreading** qualitative craft against domain standards, verifying **proof as evidence** against the original requirements and completion notes, and compiling a **pass** readiness briefing that catches the bulk of qualitative drift upfront while flagging areas that require human judgment or sensory modalities the runtime lacks.
|
|
11
11
|
|
|
12
12
|
## Operational Sequence
|
|
13
13
|
|
|
@@ -19,7 +19,7 @@ Execute this two-stage sequence in a fresh context window (or an isolated sub-ag
|
|
|
19
19
|
* Isolate **Modality & Judgment Blind Spots** - specific areas the runtime knows require verification but cannot evaluate or form a valid opinion on (for example, playtesting the tactile feel of a game mechanic, judging rendered visual aesthetics in a non-vision model, or weighing subjective brand resonance).
|
|
20
20
|
|
|
21
21
|
2. **Stage 2: State Briefing Output & Pre-Human Handover**
|
|
22
|
-
* Remediate or flag the high-confidence
|
|
22
|
+
* Remediate or flag the bulk of high-confidence craft drift without pedantic self-nitpicking, then compile the **State Briefing & Proof Pass** using the template below to guide the human's manual review session.
|
|
23
23
|
* **Filesystem vs. Read-Only Escape Hatch:** When the environment has filesystem write capabilities and saving to disk is desired, write the briefing to `state-briefing.md` (or the user's target path). When operating in an environment without write capabilities, output the entire **State Briefing & Proof Pass** directly to the user inside a fenced Markdown code block as part of your regular response.
|
|
24
24
|
|
|
25
25
|
## Proof Pass & Red-Team Guardrails
|
|
@@ -27,7 +27,7 @@ Execute this two-stage sequence in a fresh context window (or an isolated sub-ag
|
|
|
27
27
|
* **Fresh-Window Independence:** Execute `ax-proof-pass` in a context window separate from the session that planned or authored the work (either in a new chat window or via an isolated sub-agent) so the critique evaluates finished deliverables objectively rather than defending earlier compromises.
|
|
28
28
|
* **Evidence-Backed Inspection:** Rely on completion notes and outcome summaries to locate what changed quickly, inspecting the specific deliverables touched rather than burning tokens scanning unrelated materials.
|
|
29
29
|
* **Operationalize Soft-Science Taste:** Grade qualitative outputs against explicit, observable rules in the workspace's domain skills or style guides (such as sentence caps, typography tokens, contrast ratios, or structural rhythm) rather than vague subjective impressions.
|
|
30
|
-
* **
|
|
30
|
+
* **High-Signal Catch (Curb Pedantic Nitpicking):** Catch the majority of substantive qualitative drift, broken user flows, and clear guideline violations while avoiding trivial stylistic churn or endless self-correction loops.
|
|
31
31
|
* **Explicit Modality Humility:** Never fabricate opinions on sensory, experiential, or subjective qualities the runtime cannot perceive. Route every unperceivable check directly to the **Modality & Judgment Blind Spots** table so the human reviewer knows exactly where to focus.
|
|
32
32
|
|
|
33
33
|
## Stage 2 Output Template
|
|
@@ -12,33 +12,41 @@ Act as a Principal Software Engineer maintaining a codebase for both human engin
|
|
|
12
12
|
## Core Engineering Standards
|
|
13
13
|
|
|
14
14
|
### 1. Short, Functional Files
|
|
15
|
+
|
|
15
16
|
* Keep every source file short and focused so agents load minimal context per read, parse intent accurately, and discover reusable entrypoints quickly.
|
|
16
17
|
* Enforce a team-chosen line ceiling - typically 200 to 350 lines depending on language syntax verbosity - as a concrete lint rule.
|
|
17
18
|
* When a file approaches the team ceiling, decompose its logic into smaller modules before adding new behavior.
|
|
18
19
|
|
|
19
20
|
### 2. One Function Per File
|
|
20
|
-
|
|
21
|
-
*
|
|
21
|
+
|
|
22
|
+
* Export one primary function, procedure, or domain operation per file so reusable logic is never obscured or braided into a single domain flow inside multi-function utility modules.
|
|
23
|
+
* Name each file after its exported operation so directory listings act as an explicit, searchable index of available capabilities and human reviewers do not miss untested logic during top-level code coverage audits.
|
|
24
|
+
* While single-function files introduce a modest token and tool-call cost when traversing modules, pay that trade-off deliberately for long-term maintainability, test visibility, and agent discoverability.
|
|
22
25
|
|
|
23
26
|
### 3. DRY & Prefer Common Code
|
|
27
|
+
|
|
24
28
|
* Before writing new logic, search the workspace for an existing helper that already solves the problem or a close variant.
|
|
25
29
|
* When an existing helper almost fits a new call site, lightly extend its parameters or options without changing the core spirit or contract of the original helper.
|
|
26
30
|
* When a new requirement diverges from the helper's core responsibility, create a dedicated single-function module instead of overloading the existing one.
|
|
27
31
|
|
|
28
32
|
### 4. Functional Core / Imperative Shell (FCIS)
|
|
33
|
+
|
|
29
34
|
* Isolate pure data transformations, validation, parsing, and domain calculations inside deterministic core modules with zero side effects.
|
|
30
35
|
* Keep I/O, network calls, database mutations, filesystem access, and framework wiring in a thin imperative shell that delegates decisions to the functional core.
|
|
31
36
|
|
|
32
37
|
### 5. Zero Dead Code
|
|
38
|
+
|
|
33
39
|
* When a refactor changes a signature, route, or contract, delete orphaned files, unused exports, unreachable branches, and stale imports in the same turn.
|
|
34
40
|
* Never leave commented-out code blocks or unused fallback functions behind for hypothetical future use.
|
|
35
41
|
|
|
36
42
|
### 6. Strong Typing & Explicit Domain Contracts
|
|
43
|
+
|
|
37
44
|
* When starting a project or choosing a toolchain, prefer a strongly typed language or enable the strictest static type-checking mode available.
|
|
38
45
|
* Ban untyped escape hatches and implicit weak-typing fallbacks at module boundaries.
|
|
39
46
|
* Model domain inputs, outputs, and state transitions with explicit types and discriminated error or result contracts so agents receive immediate compile-time feedback, traverse definitions accurately, and prevent runtime type errors.
|
|
40
47
|
|
|
41
48
|
### 7. Actionable Error Responses
|
|
49
|
+
|
|
42
50
|
* When an application operation fails, return an explicit, concise, actionable error response rather than a generic status code or silent fallback.
|
|
43
51
|
* Include three fields in every failure payload or error message: what failed, why validation or execution rejected it, and the concrete next step to fix the request.
|
|
44
52
|
* Treat error messages as a primary API surface for both human debugging and AI agents consuming your endpoints, because actionable feedback keeps automated callers on track without guesswork.
|