@daochild/agents-config 1.2.1 → 2.2.0
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/.agents/skills/security-audit/AI-AND-LLM.md +83 -0
- package/.agents/skills/security-audit/ATTACK-CLASSES.md +130 -0
- package/.agents/skills/security-audit/CLIENT-SIDE.md +83 -0
- package/.agents/skills/security-audit/CLOUD-AND-DEPLOYMENT.md +86 -0
- package/.agents/skills/security-audit/DATA-ISOLATION-AND-LIFECYCLE.md +84 -0
- package/.agents/skills/security-audit/DESKTOP-MOBILE-AND-LOCAL-IPC.md +89 -0
- package/.agents/skills/security-audit/HUNTING.md +251 -0
- package/.agents/skills/security-audit/MEMORY-SAFETY-AND-BINARY.md +101 -0
- package/.agents/skills/security-audit/PROTOCOLS-RPC-AND-MESSAGING.md +81 -0
- package/.agents/skills/security-audit/RECONNAISSANCE.md +156 -0
- package/.agents/skills/security-audit/RESOURCE-EXHAUSTION-AND-AVAILABILITY.md +78 -0
- package/.agents/skills/security-audit/SKILL.md +192 -0
- package/.agents/skills/security-audit/SUPPLY-CHAIN-AND-RELEASE.md +73 -0
- package/.agents/skills/security-audit/VALIDATION-AND-REPORTING.md +186 -0
- package/.agents/skills/security-audit/WEB-PROTOCOL-AND-AUTH.md +105 -0
- package/.agents/skills/security-audit/report-schema.json +461 -0
- package/.agents/skills/security-audit/validate-coverage-ledger.cjs +872 -0
- package/.agents/skills/security-audit/validate-coverage-ledger.test.cjs +740 -0
- package/.agents/skills/security-audit/validate-findings.cjs +773 -0
- package/.agents/skills/security-audit/validate-findings.test.cjs +652 -0
- package/.aiignore +2 -0
- package/.github/copilot-instructions.md +19 -0
- package/AGENTS.md +85 -85
- package/ARCHITECTURE.md +37 -52
- package/CLAUDE.md +1 -0
- package/CONTRIBUTING.md +39 -24
- package/README.md +74 -2
- package/SECURITY.md +8 -0
- package/docs/BUSINESS_LOGIC.md +7 -1
- package/docs/IMPLEMENTATION.md +38 -13
- package/docs/adr/0003-document-external-skill.md +29 -0
- package/docs/stories/harness-engineering-assumptions.md +44 -0
- package/docs/stories/harness-engineering-implementation-plan.md +151 -0
- package/instructions/agent/docs.md +17 -0
- package/instructions/agent/harness-orchestrator.md +60 -0
- package/instructions/agent/harness-verifier.md +30 -0
- package/{.agents/agents → instructions/agent}/sdlc-orchestrator.md +8 -0
- package/instructions/command/harness.md +13 -0
- package/instructions/rules/harness-engineering.md +19 -0
- package/instructions/rules/nvidia-skillspector-check.md +55 -0
- package/{.claude → instructions}/rules/sdlc-gates.md +1 -1
- package/{.opencode/skill → instructions/skills}/find-skills/SKILL.md +16 -1
- package/instructions/skills/grill-me/SKILL.md +148 -0
- package/instructions/skills/harness-engineering/SKILL.md +76 -0
- package/opencode.json +78 -19
- package/package.json +20 -10
- package/scripts/sdlc-session-check.js +526 -0
- package/skills-lock.json +11 -0
- package/.agents/agents/.gitkeep +0 -0
- package/.agents/commands/.gitkeep +0 -0
- package/.agents/rules/.gitkeep +0 -0
- package/.agents/rules/sdlc-gates.md +0 -45
- package/.agents/skills/.gitkeep +0 -0
- package/.agents/skills/agents-config/SKILL.md +0 -32
- package/.claude/.gitkeep +0 -0
- package/.claude/CLAUDE.md +0 -19
- package/.claude/agents/.gitkeep +0 -0
- package/.claude/agents/docs.md +0 -15
- package/.claude/agents/sdlc-analyst.md +0 -29
- package/.claude/agents/sdlc-architect.md +0 -31
- package/.claude/agents/sdlc-auditor.md +0 -44
- package/.claude/agents/sdlc-dev.md +0 -33
- package/.claude/agents/sdlc-orchestrator.md +0 -69
- package/.claude/agents/sdlc-pm.md +0 -32
- package/.claude/agents/sdlc-qa.md +0 -33
- package/.claude/agents/sdlc-reviewer.md +0 -31
- package/.claude/agents/sdlc-security.md +0 -36
- package/.claude/agents/sdlc-sm.md +0 -31
- package/.claude/commands/.gitkeep +0 -0
- package/.claude/commands/sdlc.md +0 -18
- package/.claude/rules/.gitkeep +0 -0
- package/.claude/rules/programming-best-practices.md +0 -18
- package/.claude/skills/.gitkeep +0 -0
- package/.claude/skills/README.md +0 -3
- package/.claude/skills/sdlc-regulatory/SKILL.md +0 -515
- package/.opencode/.gitkeep +0 -0
- package/.opencode/AGENTS.md +0 -34
- package/.opencode/agent/.gitkeep +0 -0
- package/.opencode/agent/docs.md +0 -15
- package/.opencode/agent/sdlc-analyst.md +0 -30
- package/.opencode/agent/sdlc-architect.md +0 -32
- package/.opencode/agent/sdlc-auditor.md +0 -45
- package/.opencode/agent/sdlc-dev.md +0 -34
- package/.opencode/agent/sdlc-orchestrator.md +0 -70
- package/.opencode/agent/sdlc-pm.md +0 -33
- package/.opencode/agent/sdlc-qa.md +0 -34
- package/.opencode/agent/sdlc-reviewer.md +0 -32
- package/.opencode/agent/sdlc-security.md +0 -37
- package/.opencode/agent/sdlc-sm.md +0 -32
- package/.opencode/command/.gitkeep +0 -0
- package/.opencode/command/sdlc.md +0 -18
- package/.opencode/plugin/.gitkeep +0 -0
- package/.opencode/skill/.gitkeep +0 -0
- package/.opencode/skill/sdlc-regulatory/SKILL.md +0 -515
- package/docs/adr/0001-triple-mirror.md +0 -81
- package/docs/adr/0002-zero-edit-install.md +0 -237
- /package/{.agents → docs/adr}/.gitkeep +0 -0
- /package/{.agents/agents → instructions/agent}/sdlc-analyst.md +0 -0
- /package/{.agents/agents → instructions/agent}/sdlc-architect.md +0 -0
- /package/{.agents/agents → instructions/agent}/sdlc-auditor.md +0 -0
- /package/{.agents/agents → instructions/agent}/sdlc-dev.md +0 -0
- /package/{.agents/agents → instructions/agent}/sdlc-pm.md +0 -0
- /package/{.agents/agents → instructions/agent}/sdlc-qa.md +0 -0
- /package/{.agents/agents → instructions/agent}/sdlc-reviewer.md +0 -0
- /package/{.agents/agents → instructions/agent}/sdlc-security.md +0 -0
- /package/{.agents/agents → instructions/agent}/sdlc-sm.md +0 -0
- /package/{.agents/commands → instructions/command}/sdlc.md +0 -0
- /package/{.opencode/skill → instructions/skills}/archivarius/SKILL.md +0 -0
- /package/{.opencode/skill → instructions/skills}/data-engineer/SKILL.md +0 -0
- /package/{.opencode/skill → instructions/skills}/frontend-design/SKILL.md +0 -0
- /package/{.agents → instructions}/skills/sdlc-regulatory/SKILL.md +0 -0
- /package/{.opencode/skill → instructions/skills}/senior-bitcoin-auditor/SKILL.md +0 -0
- /package/{.opencode/skill → instructions/skills}/senior-frontend-dev/SKILL.md +0 -0
- /package/{.opencode/skill → instructions/skills}/senior-qa/SKILL.md +0 -0
- /package/{.opencode/skill → instructions/skills}/senior-software-developer/SKILL.md +0 -0
- /package/{.opencode/skill → instructions/skills}/senior-solidity-auditor/SKILL.md +0 -0
- /package/{.opencode/skill → instructions/skills}/senior-ux-ui-designer/SKILL.md +0 -0
package/AGENTS.md
CHANGED
|
@@ -1,14 +1,11 @@
|
|
|
1
1
|
# agents-config
|
|
2
2
|
|
|
3
|
-
A shared configuration repository for AI coding agents.
|
|
4
|
-
`.claude/` (Claude Code) and `.opencode/` (opencode) are organized and how they reference each other.
|
|
3
|
+
A shared configuration repository for AI coding agents.
|
|
5
4
|
|
|
6
5
|
## Primary instruction
|
|
7
6
|
|
|
8
|
-
Primary instructions: `node_modules/@daochild/agents-config/AGENTS.md` — read in full and follow unless overridden below.
|
|
9
|
-
|
|
10
7
|
Mandatory start-of-session checklist:
|
|
11
|
-
1. Read
|
|
8
|
+
1. Read the root AGENTS.md completely.
|
|
12
9
|
2. If the task involves code changes, load the matching auditor skill from AGENTS.md's Audit Rule before implementation.
|
|
13
10
|
3. Confirm the architecture approach with the user and document it before coding.
|
|
14
11
|
4. Do not begin implementation until steps 1–3 are satisfied.
|
|
@@ -32,8 +29,8 @@ Instructions:
|
|
|
32
29
|
## Audit Rule
|
|
33
30
|
|
|
34
31
|
When the current task involves significant code changes — for example, after the first code-generation prompt or after
|
|
35
|
-
substantial deviations from the agreed
|
|
36
|
-
output.
|
|
32
|
+
substantial deviations from the agreed statement of work (requirement/PRD artifacts; `SOW.md` was superseded by the
|
|
33
|
+
`sdlc-regulatory` skill) — the agent MUST run a system audit before finalizing the output.
|
|
37
34
|
|
|
38
35
|
1. **Pick the right auditor skill** for the domain of the change. Examples (replace `xxx` with the actual domain):
|
|
39
36
|
- Smart contracts / EVM → `senior-solidity-auditor`
|
|
@@ -42,9 +39,8 @@ output.
|
|
|
42
39
|
- General architecture / system design → `senior-software-architect`
|
|
43
40
|
2. **Run the audit via the matching `xxx-auditor` skill.** Surface findings, severity, and concrete remediation steps.
|
|
44
41
|
3. **If the required skill does not exist locally, create it first:**
|
|
45
|
-
- Add
|
|
46
|
-
-
|
|
47
|
-
- Register the skill path in `opencode.json` if needed
|
|
42
|
+
- Add `instructions/skills/<xxx-auditor>/SKILL.md` — auto-discovered via `skills.paths`, no `opencode.json` edit needed
|
|
43
|
+
- Add `.claude/` / `.agents/` counterparts only when that client is in scope (see `CONTRIBUTING.md`)
|
|
48
44
|
- Then execute the audit using the newly created skill.
|
|
49
45
|
|
|
50
46
|
Do not skip the audit because a skill is missing; create the skill and perform the audit as part of the same session.
|
|
@@ -57,13 +53,16 @@ for any feature/change that touches multiple modules or is medium/high risk.
|
|
|
57
53
|
|
|
58
54
|
### Components
|
|
59
55
|
|
|
60
|
-
| Component | Path
|
|
61
|
-
|
|
62
|
-
| Methodology skill |
|
|
63
|
-
| Always-on rule |
|
|
64
|
-
| Orchestrator subagent |
|
|
65
|
-
| Role subagents (×9) |
|
|
66
|
-
| `/sdlc` slash command |
|
|
56
|
+
| Component | Path | OpenCode registration |
|
|
57
|
+
|-----------------------|---------------------------------------------------------------------|----------------------------|
|
|
58
|
+
| Methodology skill | `instructions/skills/sdlc-regulatory/SKILL.md` | auto via `skills.paths` |
|
|
59
|
+
| Always-on rule | `instructions/rules/sdlc-gates.md` | `instructions` array |
|
|
60
|
+
| Orchestrator subagent | `instructions/agent/sdlc-orchestrator.md` | `agent` + `command.sdlc` |
|
|
61
|
+
| Role subagents (×9) | `instructions/agent/sdlc-{analyst,pm,architect,sm,dev,qa,security,reviewer,auditor}.md` | `agent` entries |
|
|
62
|
+
| `/sdlc` slash command | `instructions/command/sdlc.md` | `command.sdlc.template` |
|
|
63
|
+
|
|
64
|
+
No `.claude/` or `.agents/` mirrors exist for these today; create counterparts only when another client is in scope
|
|
65
|
+
(see `CONTRIBUTING.md`).
|
|
67
66
|
|
|
68
67
|
### SDLC chain
|
|
69
68
|
|
|
@@ -95,85 +94,86 @@ plugins, or commands.
|
|
|
95
94
|
|
|
96
95
|
```
|
|
97
96
|
agents-config/
|
|
98
|
-
├── AGENTS.md
|
|
99
|
-
├──
|
|
100
|
-
├──
|
|
101
|
-
├──
|
|
102
|
-
├──
|
|
103
|
-
├──
|
|
104
|
-
├──
|
|
105
|
-
│ ├──
|
|
106
|
-
│ ├──
|
|
107
|
-
│
|
|
108
|
-
|
|
109
|
-
├──
|
|
110
|
-
|
|
111
|
-
|
|
112
|
-
|
|
113
|
-
│ ├── settings.local.json # Personal permissions (gitignored)
|
|
114
|
-
│ ├── rules/ # Always-loaded rule files
|
|
115
|
-
│ │ ├── code-style.md
|
|
116
|
-
│ │ ├── testing.md
|
|
117
|
-
│ │ └── api-conventions.md
|
|
118
|
-
│ ├── agents/ # Subagents (Claude Code .md format)
|
|
119
|
-
│ │ ├── code-reviewer.md
|
|
120
|
-
│ │ └── security-auditor.md
|
|
121
|
-
│ ├── commands/ # Slash commands
|
|
122
|
-
│ │ ├── review.md
|
|
123
|
-
│ │ ├── fix-issue.md
|
|
124
|
-
│ │ └── deploy.md
|
|
125
|
-
│ └── skills/ # Skill folders, each with SKILL.md
|
|
126
|
-
│ ├── security-review/
|
|
127
|
-
│ │ └── SKILL.md
|
|
128
|
-
│ └── deploy/
|
|
129
|
-
│ └── SKILL.md
|
|
130
|
-
└── .opencode/ # opencode project config
|
|
131
|
-
├── AGENTS.md # opencode-scoped instructions (loaded via opencode.json)
|
|
132
|
-
├── agent/ # Subagents (opencode .md format)
|
|
133
|
-
├── skill/ # Skill folders, each with SKILL.md
|
|
134
|
-
├── command/ # Slash commands
|
|
135
|
-
└── plugin/ # Local plugins (*.ts / *.js)
|
|
97
|
+
├── AGENTS.md / ARCHITECTURE.md / CONTRIBUTING.md / README.md / SECURITY.md
|
|
98
|
+
├── CLAUDE.md # Claude Code entry: @AGENTS.md
|
|
99
|
+
├── opencode.json # instructions, skills.paths, commands, agents (schema-pinned)
|
|
100
|
+
├── package.json # scripts: test, sdlc:check, postinstall
|
|
101
|
+
├── skills-lock.json # provenance for externally sourced skills
|
|
102
|
+
├── instructions/ # OpenCode-facing sources — the real component tree
|
|
103
|
+
│ ├── agent/ # subagent prompts (sdlc-*, harness-*, docs)
|
|
104
|
+
│ ├── command/ # slash-command templates (sdlc, harness)
|
|
105
|
+
│ ├── rules/ # always-on rules (sdlc-gates, harness-engineering, nvidia-skillspector-check)
|
|
106
|
+
│ └── skills/ # first-party skills (auto-discovered)
|
|
107
|
+
├── .agents/skills/security-audit/ # external skill (ADR 0003), not in skills.paths
|
|
108
|
+
├── scripts/ # postinstall.js, sdlc-session-check.js
|
|
109
|
+
├── test/ # node --test suites, incl. opencode.json path validation
|
|
110
|
+
├── .github/copilot-instructions.md
|
|
111
|
+
└── docs/ # BUSINESS_LOGIC, IMPLEMENTATION, adr/, stories/, ...
|
|
136
112
|
```
|
|
137
113
|
|
|
138
114
|
## Linking
|
|
139
115
|
|
|
140
|
-
- `
|
|
141
|
-
|
|
142
|
-
`
|
|
143
|
-
|
|
144
|
-
|
|
145
|
-
|
|
146
|
-
|
|
147
|
-
- `.agents/` is
|
|
148
|
-
|
|
149
|
-
|
|
150
|
-
|
|
151
|
-
|
|
152
|
-
|
|
153
|
-
|
|
154
|
-
`
|
|
155
|
-
-
|
|
156
|
-
|
|
116
|
+
- OpenCode loads `opencode.json` → `instructions`: `AGENTS.md` plus three always-on rules
|
|
117
|
+
(`instructions/rules/sdlc-gates.md`, `instructions/rules/harness-engineering.md`,
|
|
118
|
+
`instructions/rules/nvidia-skillspector-check.md`). Skills are discovered only from `skills.paths` →
|
|
119
|
+
`instructions/skills`.
|
|
120
|
+
- Every `agent.*.prompt`, `command.*.template`, and `instructions` entry is a real path that `npm test` verifies —
|
|
121
|
+
if you move or rename a file under `instructions/`, update `opencode.json` in the same change.
|
|
122
|
+
- Claude Code reads root `CLAUDE.md` (`@AGENTS.md`); Copilot reads `.github/copilot-instructions.md`.
|
|
123
|
+
- `.agents/skills/security-audit/` is an externally sourced skill (provenance in `skills-lock.json`, ADR 0003); it is
|
|
124
|
+
deliberately NOT in `skills.paths`.
|
|
125
|
+
- Config loads once at startup — restart OpenCode after editing `opencode.json` or any file under `instructions/`.
|
|
126
|
+
|
|
127
|
+
## Commands
|
|
128
|
+
|
|
129
|
+
- `npm test` — runs `node --test` over `test/` (no framework config; includes `agent-config.test.js`, which fails if
|
|
130
|
+
any `opencode.json` path points at a missing file). Run this after every config/path change.
|
|
131
|
+
- `node --test --test-name-pattern="<name>" test/<file>.test.js` — run a single test.
|
|
132
|
+
- `npm run sdlc:check [-- --model provider/model]` — read-only SDLC session diagnostic; needs the OpenCode CLI and an
|
|
133
|
+
authenticated provider (see `README.md`).
|
|
134
|
+
- There is no lint/typecheck/build step — this package ships Markdown + JSON only.
|
|
157
135
|
|
|
158
136
|
## Adding a new component
|
|
159
137
|
|
|
160
|
-
|
|
161
|
-
|
|
162
|
-
|
|
163
|
-
|
|
|
164
|
-
|
|
165
|
-
|
|
|
166
|
-
|
|
|
138
|
+
Add OpenCode-facing components under `instructions/` and register them in `opencode.json` where required
|
|
139
|
+
(CONTRIBUTING.md has the full workflow, mirror rules, and frontmatter differences):
|
|
140
|
+
|
|
141
|
+
| Component | Path | Registered in `opencode.json` |
|
|
142
|
+
|----------------|-------------------------------------|------------------------------------------------|
|
|
143
|
+
| Subagent | `instructions/agent/<n>.md` | `agent.<n>.prompt` = `{file:./instructions/agent/<n>.md}` |
|
|
144
|
+
| Skill | `instructions/skills/<n>/SKILL.md` | nothing — auto-discovered via `skills.paths` |
|
|
145
|
+
| Slash command | `instructions/command/<n>.md` | `command.<n>.template` = `{file:./instructions/command/<n>.md}` |
|
|
146
|
+
| Always-on rule | `instructions/rules/<n>.md` | append to `instructions` array |
|
|
147
|
+
| Plugin | `plugin/<n>.ts` | `plugin` array (no plugins exist yet) |
|
|
148
|
+
|
|
149
|
+
`.claude/` and `.agents/` counterparts are optional and currently absent; create them only for clients in scope.
|
|
150
|
+
|
|
151
|
+
When adding a skill under `instructions/skills/`, follow `instructions/rules/nvidia-skillspector-check.md`: run NVIDIA
|
|
152
|
+
Skillspector on the skill, resolve reported issues, and attach approval artifacts to the PR. The rule is always-on for
|
|
153
|
+
OpenCode; it does not apply to end-user installs via `npx skills` outside this repository. Record external skills in
|
|
154
|
+
`skills-lock.json` when they are not on `skills.paths` (see `.agents/skills/security-audit/`).
|
|
155
|
+
|
|
156
|
+
## Harness engineering
|
|
157
|
+
|
|
158
|
+
The `harness-engineering` rule (`instructions/rules/harness-engineering.md`, always-on) and skill
|
|
159
|
+
(`instructions/skills/harness-engineering/SKILL.md`) provide shared guidance
|
|
160
|
+
for reliable agent workflows. The `/harness` command runs `harness-orchestrator`, which delegates only to
|
|
161
|
+
`general`, `sdlc-orchestrator`, and the read-only `harness-verifier`; `experimental.subagent_depth: 3` is what allows
|
|
162
|
+
harness → SDLC → role nesting. Load the skill when designing or changing agent context,
|
|
163
|
+
tools, permissions, approvals, verification, or telemetry. The consuming agent host—not
|
|
164
|
+
these configuration files—must enforce authorization, isolation, approvals, and runtime
|
|
165
|
+
limits. These properties are pinned by `test/agent-config.test.js`.
|
|
167
166
|
|
|
168
167
|
## Code Generation Pipeline
|
|
169
168
|
|
|
170
169
|
To create an agent for code generation tasks:
|
|
171
170
|
|
|
172
|
-
1. **Define the agent** in
|
|
171
|
+
1. **Define the agent** in `instructions/agent/code-generator.md` (use `mode: subagent` — `test/agent-config.test.js`
|
|
172
|
+
asserts every agent registered in `opencode.json` uses it):
|
|
173
173
|
```markdown
|
|
174
174
|
---
|
|
175
175
|
description: Generates code based on specifications and requirements
|
|
176
|
-
mode:
|
|
176
|
+
mode: subagent
|
|
177
177
|
model: anthropic/claude-sonnet-4-6
|
|
178
178
|
permission:
|
|
179
179
|
edit: ask
|
|
@@ -183,12 +183,12 @@ To create an agent for code generation tasks:
|
|
|
183
183
|
You are a code generation agent specialized in creating high-quality code...
|
|
184
184
|
```
|
|
185
185
|
|
|
186
|
-
2. **Create supporting rules** in
|
|
186
|
+
2. **Create supporting rules** in `instructions/rules/code-generation.md` and register them in the `opencode.json` `instructions` array:
|
|
187
187
|
- Coding standards and patterns
|
|
188
188
|
- Language-specific guidelines
|
|
189
189
|
- Testing requirements
|
|
190
190
|
|
|
191
|
-
3. **Add skills** in
|
|
191
|
+
3. **Add skills** in `instructions/skills/code-gen/SKILL.md`:
|
|
192
192
|
```markdown
|
|
193
193
|
---
|
|
194
194
|
name: code-gen
|
|
@@ -200,7 +200,7 @@ To create an agent for code generation tasks:
|
|
|
200
200
|
When generating code, follow these steps...
|
|
201
201
|
```
|
|
202
202
|
|
|
203
|
-
4. **Set up commands** in
|
|
203
|
+
4. **Set up commands** in `instructions/command/generate.md`:
|
|
204
204
|
```markdown
|
|
205
205
|
---
|
|
206
206
|
description: Generate code from specifications
|
|
@@ -366,7 +366,7 @@ At the end of each session or task, review and update relevant documentation:
|
|
|
366
366
|
|
|
367
367
|
### Documentation Checklist (End of Session)
|
|
368
368
|
|
|
369
|
-
- [ ] New files/directories documented in appropriate README or
|
|
369
|
+
- [ ] New files/directories documented in appropriate README or ARCHITECTURE.md
|
|
370
370
|
- [ ] Architecture decisions recorded in ADRs or ARCHITECTURE.md
|
|
371
371
|
- [ ] API changes reflected in documentation
|
|
372
372
|
- [ ] Configuration changes documented
|
package/ARCHITECTURE.md
CHANGED
|
@@ -15,54 +15,15 @@ Published to npm as `@daochild/agents-config` and intended to be installed as a
|
|
|
15
15
|
point their `AGENTS.md` at
|
|
16
16
|
`node_modules/@daochild/agents-config/AGENTS.md` to inherit this repo's conventions.
|
|
17
17
|
|
|
18
|
-
## System diagram
|
|
19
|
-
|
|
20
|
-
```
|
|
21
|
-
Consuming project
|
|
22
|
-
│
|
|
23
|
-
│ AGENTS.md → "Primary instructions are located at:
|
|
24
|
-
│ node_modules/@daochild/agents-config/AGENTS.md"
|
|
25
|
-
▼
|
|
26
|
-
agents-config (this repo)
|
|
27
|
-
│
|
|
28
|
-
├── AGENTS.md .................. single source of truth (layout + conventions)
|
|
29
|
-
│
|
|
30
|
-
├── opencode.json .............. opencode project config
|
|
31
|
-
│ ├─ instructions[] ....... always-loaded files (AGENTS.md, rules)
|
|
32
|
-
│ ├─ skills.paths[] ....... skill discovery dirs
|
|
33
|
-
│ └─ agent{} .............. subagent registry ({file:./.opencode/agent/<n>.md})
|
|
34
|
-
│
|
|
35
|
-
├── .opencode/ ................. opencode-native config
|
|
36
|
-
│ ├─ agent/ ............... subagent definitions (opencode frontmatter, mode: subagent)
|
|
37
|
-
│ ├─ skill/ ............... skills (each <n>/SKILL.md)
|
|
38
|
-
│ ├─ command/ ............. slash commands
|
|
39
|
-
│ └─ plugin/ ............. local plugins (*.ts/*.js)
|
|
40
|
-
│
|
|
41
|
-
├── .claude/ .................. Claude Code-native config
|
|
42
|
-
│ ├─ CLAUDE.md ............ Claude instructions (@AGENTS.md)
|
|
43
|
-
│ ├─ agents/ .............. subagents (Claude Code frontmatter — NO mode: field)
|
|
44
|
-
│ ├─ skills/ .............. skills (auto-discovered by opencode via skills.paths)
|
|
45
|
-
│ ├─ commands/ ............ slash commands
|
|
46
|
-
│ └─ rules/ ............... always-on rules (also listed in opencode.json instructions)
|
|
47
|
-
│
|
|
48
|
-
└── .agents/ ................... cross-client fallback (Agent Skills / agentskills.io)
|
|
49
|
-
├─ agents/ .............. verbatim opencode frontmatter
|
|
50
|
-
├─ skills/ .............. verbatim SKILL.md
|
|
51
|
-
├─ commands/ ............ verbatim
|
|
52
|
-
└─ rules/ ............... verbatim
|
|
53
|
-
```
|
|
54
|
-
|
|
55
18
|
## Architectural rules
|
|
56
19
|
|
|
57
20
|
1. **`AGENTS.md` is the single source of truth.** Both opencode (via `opencode.json`
|
|
58
21
|
→ `instructions`) and Claude Code (via `.claude/CLAUDE.md` → `@AGENTS.md`) read it. Never duplicate layout/convention
|
|
59
22
|
content in tool-specific files; reference it.
|
|
60
|
-
2. **
|
|
61
|
-
`.
|
|
62
|
-
|
|
63
|
-
|
|
64
|
-
"Adding a new component" for the per-format differences (e.g. Claude Code subagents omit the `mode:` frontmatter
|
|
65
|
-
field).
|
|
23
|
+
2. **Shared instruction sources.** OpenCode-facing agents, commands, rules, and skills are organized under
|
|
24
|
+
`instructions/{agent,command,rules,skills}/` and registered through `opencode.json`. Add `.claude/` and `.agents/`
|
|
25
|
+
counterparts for clients that need them, following the per-format differences in `AGENTS.md` (for example, Claude
|
|
26
|
+
Code subagents omit the `mode:` frontmatter field). Keep the source and any maintained counterparts synchronized.
|
|
66
27
|
3. **Skills are the methodology layer; subagents are thin.** Subagent `.md` files hold the role description,
|
|
67
28
|
responsibilities, and handoff protocol. Deep methodology (gate checklists, risk tables, compliance crosswalks) lives
|
|
68
29
|
in skills (`<n>/SKILL.md`) so it can be loaded by any agent or reviewer.
|
|
@@ -73,7 +34,7 @@ agents-config (this repo)
|
|
|
73
34
|
restart opencode. No hot reload.
|
|
74
35
|
6. **No runtime *application* code.** This repo's "code" is configuration. The only JS is the
|
|
75
36
|
opencode plugin host (`@opencode-ai/plugin` dependency), any
|
|
76
|
-
|
|
37
|
+
`plugin/*.ts|.js` the user adds, **and one constrained install-time file-pointer
|
|
77
38
|
helper (`scripts/postinstall.js`) permitted by [ADR 0002](docs/adr/0002-zero-edit-install.md)**.
|
|
78
39
|
There is no `src/`, no build output, no deployable service, no long-running process. (Rule 6
|
|
79
40
|
narrowed by ADR 0002 from "No runtime code" to "No runtime *application* code"; a single
|
|
@@ -81,24 +42,46 @@ agents-config (this repo)
|
|
|
81
42
|
|
|
82
43
|
## Key decisions
|
|
83
44
|
|
|
45
|
+
- **Dedicated harness flow (2026-10-27).** The user selected KISS: a thin
|
|
46
|
+
`harness-orchestrator` coordinates context, policy, execution, independent verification,
|
|
47
|
+
and feedback using the existing harness skill. A separate read-only `harness-verifier`
|
|
48
|
+
checks outcomes; neither role grants authority or replaces SDLC gates. The change is
|
|
49
|
+
medium risk with a 70% autonomy cap; human review and deployment remain pending.
|
|
50
|
+
Traceability id: `dedicated-harness-flow`.
|
|
51
|
+
- **Client loading and dispatch.** Use the discoverable `opencode.json` filename
|
|
52
|
+
(formerly `opencode.jsonp`) and explicitly register commands stored under
|
|
53
|
+
`instructions/command/`. Other hosts must load role instructions into supported
|
|
54
|
+
subagents; OpenCode configuration does not register named Copilot agent types.
|
|
55
|
+
Missing dispatch capability is a host limitation, not a failed engineering gate.
|
|
56
|
+
`experimental.subagent_depth: 3` permits the finite harness -> SDLC -> role
|
|
57
|
+
chain; the default single level would block it. The harness coordinator denies
|
|
58
|
+
direct side effects and restricts delegation; the verifier is read-only.
|
|
59
|
+
The nesting setting requires a compatible OpenCode version and live validation.
|
|
84
60
|
- **Schema-pinned config.** `opencode.json` declares
|
|
85
61
|
`"$schema": "https://opencode.ai/config.json"` for editor IntelliSense and validation against the opencode config
|
|
86
62
|
schema.
|
|
87
|
-
- **
|
|
88
|
-
|
|
89
|
-
|
|
90
|
-
|
|
91
|
-
|
|
92
|
-
|
|
63
|
+
- **Client-specific frontmatter.** OpenCode agent files use `mode: subagent` (and optional `permission`). Claude Code
|
|
64
|
+
agent files omit `mode:`. `.agents/` counterparts retain OpenCode frontmatter.
|
|
65
|
+
- **Explicit skill discovery.** `opencode.json` lists `instructions/skills`; keep this path aligned with the actual
|
|
66
|
+
skill source directory when changing the layout.
|
|
67
|
+
- **External skill provenance.** The imported `security-audit` Agent Skill lives under
|
|
68
|
+
`.agents/skills/security-audit/`; `skills-lock.json` records its upstream source and computed hash. Keep it
|
|
69
|
+
documented separately from OpenCode's `instructions/skills` discovery path. See
|
|
70
|
+
[ADR 0003](docs/adr/0003-document-external-skill.md).
|
|
93
71
|
- **SDLC regulatory framework.** The `sdlc-*` subagents + `sdlc-regulatory` skill +
|
|
94
72
|
`sdlc-gates` rule implement the AI-Driven SDLC. The skill is the source of truth for the methodology (it superseded
|
|
95
73
|
the deleted `SOW.md`). The orchestrator is the single entry point via the `/sdlc` command.
|
|
74
|
+
- **Harness engineering guidance.** The `harness-engineering` skill and always-on rule are
|
|
75
|
+
stored under `instructions/` for OpenCode and mirrored for other clients when needed. They describe context selection,
|
|
76
|
+
host-enforced policy, bounded execution, independent verification, and privacy-aware
|
|
77
|
+
observability. They are guidance only; consuming runtimes enforce permissions, approvals,
|
|
78
|
+
isolation, and execution limits.
|
|
96
79
|
|
|
97
80
|
## What must NOT be changed without human approval
|
|
98
81
|
|
|
99
82
|
- `AGENTS.md` — layout and linking model (any change breaks both tools).
|
|
100
83
|
- `opencode.json` schema pin and `instructions` list (changes what's always-loaded).
|
|
101
|
-
- The
|
|
84
|
+
- The instruction-source layout and any maintained client counterparts; removing a counterpart can break that client.
|
|
102
85
|
- The `sdlc-regulatory` skill's gate checklists and risk-autonomy table (these are the control surface; relaxing them is
|
|
103
86
|
a high-risk change per the skill's own risk model).
|
|
104
87
|
- **The `postinstall` script's contract (per [ADR 0002](docs/adr/0002-zero-edit-install.md))**:
|
|
@@ -140,4 +123,6 @@ agents-config (this repo)
|
|
|
140
123
|
- `CONTRIBUTING.md` — how to add agents/skills/commands/rules.
|
|
141
124
|
- `SECURITY.md` — security policy and the LLM security rule.
|
|
142
125
|
- `docs/` — business logic, implementation status, architecture deep-dives, ADRs.
|
|
143
|
-
-
|
|
126
|
+
- `instructions/skills/sdlc-regulatory/SKILL.md` — AI-Driven SDLC methodology (source of truth for the SDLC controls).
|
|
127
|
+
- `instructions/skills/harness-engineering/SKILL.md` — shared agent reliability guidance; runtime enforcement remains the consuming host's responsibility.
|
|
128
|
+
- `docs/adr/0003-document-external-skill.md` — documents the source and discovery boundary for externally sourced skills.
|
package/CLAUDE.md
ADDED
|
@@ -0,0 +1 @@
|
|
|
1
|
+
@AGENTS.md
|
package/CONTRIBUTING.md
CHANGED
|
@@ -9,55 +9,70 @@ files** (agents, skills, commands, rules, plugins), not runtime code.
|
|
|
9
9
|
## Before you start
|
|
10
10
|
|
|
11
11
|
1. Read `AGENTS.md` — it is the single source of truth for layout and conventions.
|
|
12
|
-
2. Read `ARCHITECTURE.md` — explains the
|
|
12
|
+
2. Read `ARCHITECTURE.md` — explains the instruction-source layout and what must not be changed without approval.
|
|
13
13
|
3. Read `SECURITY.md` — the LLM security rule applies to every change.
|
|
14
14
|
4. If your change touches the SDLC, load the `sdlc-regulatory` skill; if it is medium/high risk or spans multiple
|
|
15
15
|
components, run it through `/sdlc`.
|
|
16
16
|
|
|
17
|
-
## Adding a component
|
|
17
|
+
## Adding a component
|
|
18
18
|
|
|
19
|
-
|
|
20
|
-
|
|
19
|
+
Add OpenCode-facing components under `instructions/` and register them in `opencode.json` where required. When a
|
|
20
|
+
component must also work in Claude Code or Agent-Skills-compatible clients, maintain the corresponding mirrors listed
|
|
21
|
+
below. Do not assume a mirror directory exists; create the required target and keep its format consistent with that
|
|
22
|
+
client.
|
|
21
23
|
|
|
22
24
|
| Component | opencode path | Claude Code path | Agent Skills path |
|
|
23
25
|
|----------------|---------------------------------------|-------------------------------|-------------------------------|
|
|
24
|
-
| Subagent |
|
|
25
|
-
| Skill |
|
|
26
|
-
| Slash command |
|
|
27
|
-
| Plugin |
|
|
28
|
-
| Always-on rule |
|
|
26
|
+
| Subagent | `instructions/agent/<n>.md` | `.claude/agents/<n>.md` | `.agents/agents/<n>.md` |
|
|
27
|
+
| Skill | `instructions/skills/<n>/SKILL.md` | `.claude/skills/<n>/SKILL.md` | `.agents/skills/<n>/SKILL.md` |
|
|
28
|
+
| Slash command | `instructions/command/<n>.md` | `.claude/commands/<n>.md` | `.agents/commands/<n>.md` |
|
|
29
|
+
| Plugin | `plugin/<n>.ts` | n/a | n/a |
|
|
30
|
+
| Always-on rule | `instructions/rules/<n>.md` | `.claude/rules/<n>.md` | `.agents/rules/<n>.md` |
|
|
31
|
+
|
|
32
|
+
### Adding a skill to this repository
|
|
33
|
+
|
|
34
|
+
Installing a skill with `npx skills` on a developer machine is separate from **landing** a skill in `agents-config`.
|
|
35
|
+
Before you add or update `instructions/skills/<n>/SKILL.md` (and any `.claude/` / `.agents/` mirrors), comply with
|
|
36
|
+
`instructions/rules/nvidia-skillspector-check.md`:
|
|
37
|
+
|
|
38
|
+
1. Run NVIDIA Skillspector on the skill source and fix compliance, security, or compatibility findings.
|
|
39
|
+
2. Pass the SDLC Implementation Gate when the change is medium/high risk or spans multiple components (`/sdlc`).
|
|
40
|
+
3. Add the skill under `instructions/skills/<n>/SKILL.md` — no `opencode.json` entry needed beyond `skills.paths`.
|
|
41
|
+
4. Put Skillspector reports or approval artifacts in the PR description.
|
|
42
|
+
5. Update `skills-lock.json` when the skill is externally sourced and not on `skills.paths`.
|
|
43
|
+
|
|
44
|
+
Use the `find-skills` skill to discover candidates from the open ecosystem; always complete Skillspector before merging
|
|
45
|
+
into this repo.
|
|
29
46
|
|
|
30
47
|
### Format differences
|
|
31
48
|
|
|
32
49
|
- **opencode subagent frontmatter** uses `mode: subagent` (and optional `permission`).
|
|
33
50
|
- **Claude Code subagent frontmatter** omits the `mode:` field.
|
|
34
|
-
- **`.agents/` mirrors** keep the
|
|
51
|
+
- **`.agents/` mirrors** keep the OpenCode frontmatter verbatim.
|
|
35
52
|
- **Skills** use the same `SKILL.md` format in all three locations.
|
|
36
53
|
|
|
37
54
|
## Workflow
|
|
38
55
|
|
|
39
56
|
1. Create a branch: `feat/<name>` or `chore/<name>`.
|
|
40
|
-
2. Make the change
|
|
41
|
-
3.
|
|
42
|
-
4. If you added/changed an always-on rule,
|
|
43
|
-
`.agents/rules/`.
|
|
57
|
+
2. Make the OpenCode-facing change under `instructions/`.
|
|
58
|
+
3. Add or update the `.claude/` counterpart (apply format differences above) and `.agents/` counterpart (verbatim) when those clients are in scope.
|
|
59
|
+
4. If you added/changed an always-on rule, register its actual path in `opencode.json` `instructions` and add the file to `.claude/rules/` and `.agents/rules/` when those clients are in scope.
|
|
44
60
|
5. If you added a subagent, register it under `agent` in `opencode.json` with a
|
|
45
|
-
`prompt: "{file
|
|
61
|
+
`prompt: "{file:./instructions/agent/<n>.md}"` entry.
|
|
46
62
|
6. **Restart opencode** to validate — config is loaded once at startup, not hot-reloaded.
|
|
47
|
-
7. Verify
|
|
63
|
+
7. Verify any maintained mirrors before committing (see the examples below).
|
|
48
64
|
|
|
49
65
|
## Verifying mirror consistency
|
|
50
66
|
|
|
51
|
-
Before committing, confirm
|
|
67
|
+
Before committing, confirm each maintained counterpart is consistent with the source:
|
|
52
68
|
|
|
53
69
|
```bash
|
|
54
|
-
#
|
|
55
|
-
diff -q
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
diff -q
|
|
60
|
-
diff -q .opencode/skill/<n>/SKILL.md .agents/skills/<n>/SKILL.md
|
|
70
|
+
# .agents keeps OpenCode frontmatter; Claude Code subagents omit mode: subagent
|
|
71
|
+
diff -q instructions/agent/<n>.md .agents/agents/<n>.md
|
|
72
|
+
diff <(grep -v '^mode: subagent$' instructions/agent/<n>.md) .claude/agents/<n>.md
|
|
73
|
+
# skills share the same content
|
|
74
|
+
diff -q instructions/skills/<n>/SKILL.md .claude/skills/<n>/SKILL.md
|
|
75
|
+
diff -q instructions/skills/<n>/SKILL.md .agents/skills/<n>/SKILL.md
|
|
61
76
|
```
|
|
62
77
|
|
|
63
78
|
## Commit conventions
|
package/README.md
CHANGED
|
@@ -7,9 +7,10 @@ The package is intended for production‑grade automation: version control, CI/C
|
|
|
7
7
|
## What's in the box
|
|
8
8
|
|
|
9
9
|
- **Shared layout & conventions** — `AGENTS.md` is the single source of truth, inherited (not copied) by consuming projects.
|
|
10
|
-
- **
|
|
10
|
+
- **Shared instruction sources** — OpenCode-facing agents, commands, rules, and skills are organized under `instructions/`; `opencode.json` declares the paths used by OpenCode. Keep configured paths aligned with the files and maintain Claude Code or Agent Skills counterparts when required; see `AGENTS.md` and `CONTRIBUTING.md`. New skills under `instructions/skills/` require NVIDIA Skillspector approval per `instructions/rules/nvidia-skillspector-check.md` (see `CONTRIBUTING.md`).
|
|
11
11
|
- **AI-Driven SDLC framework** — `sdlc-regulatory` skill + 10 `sdlc-*` subagents + `sdlc-gates` rule + `/sdlc` command. Gates run in order (requirement → architecture → implementation → testing → security → review → deployment → DoD), AI never self-validates, high-risk changes require human approval, and a compliance crosswalk covers EU AI Act / NIST AI RMF / ISO 42001 / SOC 2 / GDPR.
|
|
12
|
-
- **Domain skills** — `senior-qa`, `senior-solidity-auditor`, `senior-bitcoin-auditor`, `senior-frontend-dev`, `senior-software-developer`, `senior-ux-ui-designer`, `archivarius`, `data-engineer`, `find-skills`, `frontend-design`.
|
|
12
|
+
- **Domain skills** — `senior-qa`, `senior-solidity-auditor`, `senior-bitcoin-auditor`, `senior-frontend-dev`, `senior-software-developer`, `senior-ux-ui-designer`, `archivarius`, `data-engineer`, `find-skills`, `frontend-design`, `grill-me`.
|
|
13
|
+
- **Externally sourced skill** — `security-audit` is maintained at `.agents/skills/security-audit/` and sourced from `cloudflare/security-audit-skill`. Its source and computed hash are recorded in `skills-lock.json`. It follows the Agent Skills layout and is separate from OpenCode's `instructions/skills` discovery path.
|
|
13
14
|
- **Always-on rules** — `programming-best-practices` and `sdlc-gates`.
|
|
14
15
|
- **AI-readable contract** — `ARCHITECTURE.md`, `CONTRIBUTING.md`, `SECURITY.md`, `docs/` (business logic, implementation status, ADRs).
|
|
15
16
|
|
|
@@ -75,6 +76,77 @@ This manual path is the supported fallback and is runnable without the
|
|
|
75
76
|
Restart your agent session (opencode or Claude Code) so the new instructions
|
|
76
77
|
take effect — config is loaded once at startup and is not hot-reloaded.
|
|
77
78
|
|
|
79
|
+
## Read-only SDLC session diagnostic
|
|
80
|
+
|
|
81
|
+
### Agent flows and loading
|
|
82
|
+
|
|
83
|
+
OpenCode loads the root `opencode.json` (previously misnamed `opencode.jsonp`).
|
|
84
|
+
Restart OpenCode after configuration changes. `/sdlc <requirement>` and
|
|
85
|
+
`/harness <task>` are explicitly registered there because `instructions/command/`
|
|
86
|
+
is not an automatically discovered command directory.
|
|
87
|
+
The configuration sets `experimental.subagent_depth` to 3 for the bounded
|
|
88
|
+
harness -> SDLC orchestrator -> role chain. The current schema defaults to one
|
|
89
|
+
level, which prevents subagent orchestrators from dispatching their children.
|
|
90
|
+
This setting is version-sensitive; use an OpenCode release supporting it.
|
|
91
|
+
|
|
92
|
+
`/harness` runs context -> policy -> execution -> independent verification ->
|
|
93
|
+
feedback. It uses the existing harness skill, delegates implementation through
|
|
94
|
+
the SDLC when required, and uses a separate read-only `harness-verifier`.
|
|
95
|
+
The consuming host must enforce permissions, approvals, isolation, and finite
|
|
96
|
+
execution budgets; configuration prose alone cannot provide these controls.
|
|
97
|
+
The harness coordinator denies direct execution and delegates only to `general`,
|
|
98
|
+
`sdlc-orchestrator`, and `harness-verifier`; the verifier denies execution and
|
|
99
|
+
delegation. These OpenCode permissions do not configure other agent hosts.
|
|
100
|
+
|
|
101
|
+
**If SDLC is blocked:** distinguish host capability from gate failure.
|
|
102
|
+
OpenCode agent registrations do not create named agent types in Copilot or
|
|
103
|
+
other hosts. Where supported, dispatch a general-purpose subagent with the
|
|
104
|
+
appropriate `instructions/agent/sdlc-<role>.md` instructions and keep developer,
|
|
105
|
+
security reviewer, reviewer, and auditor in distinct contexts. If the host
|
|
106
|
+
cannot dispatch independent agents, request human review instead of claiming
|
|
107
|
+
a gate passed. Missing approvals remain pending; missing deployment evidence
|
|
108
|
+
prevents production/DoD sign-off, not earlier authorized local work.
|
|
109
|
+
|
|
110
|
+
`npm run sdlc:check` intentionally denies subagents and shell commands. It checks
|
|
111
|
+
a read-only response, not the full SDLC flow; do not use it to prove role dispatch.
|
|
112
|
+
|
|
113
|
+
### Running the diagnostic
|
|
114
|
+
|
|
115
|
+
Run a small documentation-analysis task through OpenCode and check whether the
|
|
116
|
+
response addresses the SDLC gates without changing the repository:
|
|
117
|
+
|
|
118
|
+
```bash
|
|
119
|
+
npm run sdlc:check
|
|
120
|
+
# Optional: select another configured model, including a provider's free model
|
|
121
|
+
npm run sdlc:check -- --model provider/model
|
|
122
|
+
```
|
|
123
|
+
|
|
124
|
+
The default model candidate is `opencode/beam-free`; model availability and
|
|
125
|
+
provider requirements can change, so use `--model` for a currently available
|
|
126
|
+
free model. OpenCode CLI and an authenticated provider are required.
|
|
127
|
+
|
|
128
|
+
The runner injects explicit read-only OpenCode permissions, denies shell and
|
|
129
|
+
subagent access, and hashes repository files before and after the session
|
|
130
|
+
(including ignored files, excluding `.git`). It writes one private
|
|
131
|
+
session directory under `~/.local/state/agents-config/sdlc-session-checks/`
|
|
132
|
+
containing `events.jsonl` and `summary.json`. These logs contain tool names,
|
|
133
|
+
statuses, observed SDLC subagents, and gate-mention checks only; they do not
|
|
134
|
+
store the prompt, response, tool arguments, or tool outputs. OpenCode itself
|
|
135
|
+
may retain its normal session history according to its own configuration.
|
|
136
|
+
|
|
137
|
+
`Assessment: PASS` means the diagnostic task completed, no file-write tool or
|
|
138
|
+
other disallowed tool was attempted, repository contents stayed unchanged, and
|
|
139
|
+
the response reported a risk classification plus a status for every gate. A
|
|
140
|
+
blocked shell attempt still makes the assessment fail, but the file snapshot
|
|
141
|
+
shows whether anything changed. This is a read-only smoke check, not proof that
|
|
142
|
+
an agent obeys policy in every session or that a gate was actually
|
|
143
|
+
approved/passed. The permission policy governs OpenCode tool calls, not
|
|
144
|
+
arbitrary side effects from external plugins or host processes; use an
|
|
145
|
+
OS-level read-only sandbox when those must be excluded as well. Run tests
|
|
146
|
+
without OpenCode using `npm test`. Gate labels/statuses and risk classification
|
|
147
|
+
are heuristic checks of the response, not independent validation that its
|
|
148
|
+
claims are accurate or that the underlying gates passed.
|
|
149
|
+
|
|
78
150
|
## License
|
|
79
151
|
|
|
80
152
|
MIT
|
package/SECURITY.md
CHANGED
|
@@ -54,6 +54,13 @@ Security concerns are therefore primarily:
|
|
|
54
54
|
pinned and monitor advisories.
|
|
55
55
|
- **Permission overreach** in `opencode.json` agent `permission` blocks — prefer
|
|
56
56
|
`ask` over `allow` for `edit` and `bash` on any agent that touches production.
|
|
57
|
+
- **Model-only policy enforcement** — prompts and skills do not authorize actions or
|
|
58
|
+
isolate tools. The consuming host must check identity, scope, and arguments at each
|
|
59
|
+
side-effect boundary, fail closed when authorization is unclear, and pause for required
|
|
60
|
+
human approval.
|
|
61
|
+
- **Sensitive observability data** — agent traces can contain secrets or personal data.
|
|
62
|
+
Record only operational fields needed for diagnosis; redact sensitive values and apply
|
|
63
|
+
access and retention controls. Do not log private model reasoning by default.
|
|
57
64
|
- **SDLC gate bypass** — any change that weakens a gate, removes a reviewer, or
|
|
58
65
|
raises an autonomy cap is a high-risk security change requiring human approval.
|
|
59
66
|
|
|
@@ -75,5 +82,6 @@ lists. This policy does not replace legal review.
|
|
|
75
82
|
|
|
76
83
|
- `AGENTS.md` → "Security Guidelines"
|
|
77
84
|
- `sdlc-regulatory` skill → Security Gate checklist + compliance crosswalk
|
|
85
|
+
- `instructions/skills/harness-engineering/SKILL.md` → harness design guidance and host/runtime boundary
|
|
78
86
|
- `CONTRIBUTING.md` → SDLC changes are high-risk
|
|
79
87
|
- `.gitignore`, `.npmignore` — what is excluded from repo and package
|