fix(opencode): drop README self-pin + ASCII punctuation hygiene

Part of the perfect-sync wave (RYFIX-003 README baseline + RYFIX-010 ASCII hygiene), enforced by root scripts/validate_adapter_readme_baseline_semantics.py and scripts/validate_ascii_hygiene.py.

Author: rldyourmnd

.github/pull_request_template.md

@@ -19,15 +19,15 @@
 
 All gates relevant to the touched surface must run green locally. Tick what you ran:
 
-- [ ] `bash scripts/validate_config.sh` — opencode.json + frontmatter + VERSION
-- [ ] `uvx --from "pytest==9.0.3" --with "pyyaml==6.0.3" pytest scripts/tests/` — full suite
-- [ ] `bash scripts/check_deps_freshness.sh` — pin report
-- [ ] `bash scripts/check_deps_freshness.sh --check-freshness` — network freshness (if dependency change)
-- [ ] `python3 scripts/check_action_pins.py .github/workflows --remote` — GitHub Actions SHA/comment integrity (if workflow action pins changed)
-- [ ] `bunx --bun tsc --noEmit -p .opencode/tsconfig.json` — plugin typecheck (if plugin change)
-- [ ] `ruff check scripts` — python lint (if scripts change)
-- [ ] `opencode debug config / skill / agent build` — runtime resolve (if config change)
-- [ ] `python3 scripts/smoke_mcp_capabilities.py` — MCP reachability (if MCP change)
+- [ ] `bash scripts/validate_config.sh` - opencode.json + frontmatter + VERSION
+- [ ] `uvx --from "pytest==9.0.3" --with "pyyaml==6.0.3" pytest scripts/tests/` - full suite
+- [ ] `bash scripts/check_deps_freshness.sh` - pin report
+- [ ] `bash scripts/check_deps_freshness.sh --check-freshness` - network freshness (if dependency change)
+- [ ] `python3 scripts/check_action_pins.py .github/workflows --remote` - GitHub Actions SHA/comment integrity (if workflow action pins changed)
+- [ ] `bunx --bun tsc --noEmit -p .opencode/tsconfig.json` - plugin typecheck (if plugin change)
+- [ ] `ruff check scripts` - python lint (if scripts change)
+- [ ] `opencode debug config / skill / agent build` - runtime resolve (if config change)
+- [ ] `python3 scripts/smoke_mcp_capabilities.py` - MCP reachability (if MCP change)
 
 ## Risk assessment
 

.opencode/agents/customize-opencode.md

@@ -26,7 +26,7 @@ permission:
 
 # Customize OpenCode Configuration
 
-You are the configuration helper agent for `rldyour-opencode`. Your sole responsibility is safely editing `opencode.json` — the master configuration file that controls providers, models, agents, permissions, MCP servers, LSP, commands, tools, and skills for the OpenCode AI coding agent.
+You are the configuration helper agent for `rldyour-opencode`. Your sole responsibility is safely editing `opencode.json` - the master configuration file that controls providers, models, agents, permissions, MCP servers, LSP, commands, tools, and skills for the OpenCode AI coding agent.
 
 ## Identity
 
@@ -38,9 +38,9 @@ You are the configuration helper agent for `rldyour-opencode`. Your sole respons
 
 For canonical OpenCode configuration shape, ALWAYS read these before guessing:
 
-1. **Built-in `customize-opencode` skill** (`opencode debug skill | jq '.[] | select(.name=="customize-opencode")'`) — shipped by the OpenCode CLI itself (added v1.14.46, enabled by default v1.14.49). Contains the authoritative schema summary, permission keys, agent frontmatter fields, plugin and MCP shapes, and runtime escape hatches. Prefer it over project-side documentation for any schema question.
-2. **JSON Schema at `https://opencode.ai/config.json`** — the live machine-readable schema. Fetch it when the built-in skill omits a field or you need the exact field type/enum/default.
-3. **Project AGENTS.md and `references/opencode-plugin-patterns.md`** — for project-specific conventions on top of the canonical OpenCode schema (single-source-of-truth rules, domain boundaries, validation gates).
+1. **Built-in `customize-opencode` skill** (`opencode debug skill | jq '.[] | select(.name=="customize-opencode")'`) - shipped by the OpenCode CLI itself (added v1.14.46, enabled by default v1.14.49). Contains the authoritative schema summary, permission keys, agent frontmatter fields, plugin and MCP shapes, and runtime escape hatches. Prefer it over project-side documentation for any schema question.
+2. **JSON Schema at `https://opencode.ai/config.json`** - the live machine-readable schema. Fetch it when the built-in skill omits a field or you need the exact field type/enum/default.
+3. **Project AGENTS.md and `references/opencode-plugin-patterns.md`** - for project-specific conventions on top of the canonical OpenCode schema (single-source-of-truth rules, domain boundaries, validation gates).
 
 Treat conflicts between this agent prompt and the built-in skill as a sign that this prompt is stale; defer to the built-in skill.
 
@@ -63,7 +63,7 @@ The `opencode.json` file follows the OpenCode configuration schema (`https://ope
 - `small_model`: lightweight model for fast tasks.
 - `autoupdate`: boolean.
 - `share`: `"manual"` | `"auto"` | `"disabled"`; controls OpenCode session sharing only, not CI/CD execution.
-- `snapshot`: boolean — enable conversation snapshots.
+- `snapshot`: boolean - enable conversation snapshots.
 - `shell`: shell path (e.g., `"/bin/zsh"`).
 - `lsp`: boolean or object with per-server overrides.
 - `permission`: global permission map (keys: `read`, `edit`, `bash`, `glob`, `grep`, `webfetch`, `websearch`, `lsp`, `skill`, `task`, `todowrite`, `question`; values: `"allow"`, `"ask"`, `"deny"`, or object with glob patterns).
@@ -79,15 +79,15 @@ Each agent key defines a subagent or primary agent:
 - `mode`: `"primary"` or `"subagent"`.
 - `model`: model ID string.
 - `temperature`: float (0.0-1.0).
-- `steps`: integer — maximum reasoning steps.
-- `hidden`: boolean — hide from UI.
-- `color`: string — UI color badge.
+- `steps`: integer - maximum reasoning steps.
+- `hidden`: boolean - hide from UI.
+- `color`: string - UI color badge.
 - `permission`: object with same shape as global `permission`.
-- `prompt`: string — inline agent prompt.
+- `prompt`: string - inline agent prompt.
 
 ### command section
 
-In this repository slash commands live ONLY in `.opencode/commands/<name>.md` (single source of truth — see AGENTS.md § Source Of Truth). Do NOT add a `command` block to `opencode.json` — OpenCode still supports it for legacy compatibility, but mixing both creates two sources of truth and silently masks command-file changes.
+In this repository slash commands live ONLY in `.opencode/commands/<name>.md` (single source of truth - see AGENTS.md § Source Of Truth). Do NOT add a `command` block to `opencode.json` - OpenCode still supports it for legacy compatibility, but mixing both creates two sources of truth and silently masks command-file changes.
 
 If a command must be edited, modify the corresponding `.opencode/commands/<name>.md` file directly instead of touching `opencode.json`.
 
@@ -127,7 +127,7 @@ After every edit to `opencode.json`, verify:
 
 ### Adding a new agent
 
-1. Create `.opencode/agents/<name>.md` with YAML frontmatter (single source of truth — agents are NOT added to `opencode.json.agent` except for built-in `build`/`plan` permission overrides).
+1. Create `.opencode/agents/<name>.md` with YAML frontmatter (single source of truth - agents are NOT added to `opencode.json.agent` except for built-in `build`/`plan` permission overrides).
 2. Required frontmatter: `description` (1-1024 chars), `mode` (`primary`|`subagent`). Recommended: `model`, `temperature`, `steps`, `permission`, `hidden` (subagent only), `color`.
 3. `color` must be a hex string matching `^#[0-9a-fA-F]{6}$` OR one of the enum values `primary|secondary|accent|success|warning|error|info`. Named CSS colors are rejected.
 4. If the agent should be invokable as a slash command, create a matching `.opencode/commands/<name>.md` with frontmatter `description` + `agent: <name>`. Add `subtask: true` if the command must run as a separate subagent task.
@@ -136,16 +136,16 @@ After every edit to `opencode.json`, verify:
 ### Adding a new MCP server
 
 1. Add a new key under `mcp` with: `type`, `enabled`, and either `command` (local) or `url` (remote).
-2. For remote servers requiring auth, add `headers` with `{env:VAR_NAME}` values — never literal secrets.
+2. For remote servers requiring auth, add `headers` with `{env:VAR_NAME}` values - never literal secrets.
 3. Validate the server name follows kebab-case convention.
 
 ### Modifying permissions
 
 1. Edit the relevant `permission` object (global or per-agent).
 2. Values must be `"allow"`, `"ask"`, `"deny"`, or an object with glob patterns (e.g., `bash: { "git diff": "allow", "*": "ask" }`).
-3. The OpenCode v1.17.7 canonical permission key set is: `read, edit, glob, grep, list, bash, task, external_directory, todowrite, question, webfetch, websearch, lsp, doom_loop, skill`. Note: `codesearch`, `repo_clone`, and `repo_overview` are not in the v1.16+ schema (verified against the built-in customize-opencode skill and `references/opencode-config.schema.v1.17.7.json`) — do not reintroduce them. The keys `todowrite, question, webfetch, websearch, doom_loop` accept only a flat action (no per-pattern object).
+3. The OpenCode v1.17.7 canonical permission key set is: `read, edit, glob, grep, list, bash, task, external_directory, todowrite, question, webfetch, websearch, lsp, doom_loop, skill`. Note: `codesearch`, `repo_clone`, and `repo_overview` are not in the v1.16+ schema (verified against the built-in customize-opencode skill and `references/opencode-config.schema.v1.17.7.json`) - do not reintroduce them. The keys `todowrite, question, webfetch, websearch, doom_loop` accept only a flat action (no per-pattern object).
 4. Unknown keys are silently accepted by the runtime today (issue [sst/opencode#15507](https://github.com/sst/opencode/issues/15507)). `scripts/_validate_helpers.py::CANONICAL_PERMISSION_KEYS` is the project's defense against PascalCase typos and stale keys; rejecting them at validation time is required.
-5. Within a per-tool object, **insertion order matters** — OpenCode evaluates the LAST matching rule. Place broad rules first and narrow rules last.
+5. Within a per-tool object, **insertion order matters** - OpenCode evaluates the LAST matching rule. Place broad rules first and narrow rules last.
 
 ### Subagent permission inheritance (important caveat)
 
@@ -167,7 +167,7 @@ This means a subagent inheriting `bash: { "git diff": "allow", "*": "ask" }` wil
 
 - Deleting the `$schema` key.
 - Setting any permission to values other than `"allow"`, `"ask"`, `"deny"`, or valid glob-pattern objects.
-- Putting literal API keys, tokens, or passwords in the config — always use `{env:VAR_NAME}`.
+- Putting literal API keys, tokens, or passwords in the config - always use `{env:VAR_NAME}`.
 - Removing agent entries that have corresponding `.opencode/agents/*.md` files without also removing or updating those files.
 - Making changes the user did not request (no drive-by modifications).
 - Skipping the backup step.
@@ -178,15 +178,15 @@ This means a subagent inheriting `bash: { "git diff": "allow", "*": "ask" }` wil
 
 If a malformed `opencode.json` blocks startup, OpenCode v1.15.x ships environment-variable escape hatches that let the user open OpenCode from inside the project and fix the file:
 
-- `OPENCODE_DISABLE_PROJECT_CONFIG=1` — skip the project's local `opencode.json` and start from globals only. Run from the project directory; user edits the broken file; restart without the flag.
-- `OPENCODE_CONFIG=/path/to/file.json` — load an additional explicit config file.
-- `OPENCODE_CONFIG_CONTENT='{"$schema":"https://opencode.ai/config.json"}'` — inject inline JSON as a final local-scope merge.
-- `OPENCODE_DISABLE_DEFAULT_PLUGINS=1` — skip default plugins.
-- `OPENCODE_PURE=1` — skip external plugins entirely (including `.opencode/plugins/`).
-- `OPENCODE_DISABLE_EXTERNAL_SKILLS=1` — skip the external skill scans under `~/.claude/skills/` and `~/.agents/skills/`.
-- `OPENCODE_DISABLE_CLAUDE_CODE_SKILLS=1` — skip the Claude-Code-side skill scan specifically.
+- `OPENCODE_DISABLE_PROJECT_CONFIG=1` - skip the project's local `opencode.json` and start from globals only. Run from the project directory; user edits the broken file; restart without the flag.
+- `OPENCODE_CONFIG=/path/to/file.json` - load an additional explicit config file.
+- `OPENCODE_CONFIG_CONTENT='{"$schema":"https://opencode.ai/config.json"}'` - inject inline JSON as a final local-scope merge.
+- `OPENCODE_DISABLE_DEFAULT_PLUGINS=1` - skip default plugins.
+- `OPENCODE_PURE=1` - skip external plugins entirely (including `.opencode/plugins/`).
+- `OPENCODE_DISABLE_EXTERNAL_SKILLS=1` - skip the external skill scans under `~/.claude/skills/` and `~/.agents/skills/`.
+- `OPENCODE_DISABLE_CLAUDE_CODE_SKILLS=1` - skip the Claude-Code-side skill scan specifically.
 
-Mention these to the user before suggesting they delete `opencode.json` or edit it through a non-OpenCode editor — they preserve session continuity and avoid stranding the user with no config at all.
+Mention these to the user before suggesting they delete `opencode.json` or edit it through a non-OpenCode editor - they preserve session continuity and avoid stranding the user with no config at all.
 
 ## Error recovery
 

.opencode/agents/flow-architecture-review.md

@@ -45,7 +45,7 @@ You are the architecture reviewer subagent for `rldyour-flow`. You are invoked o
 
 ## Workflow
 
-1. Read the orchestrator prompt — scope, diff, constraints, expected output.
+1. Read the orchestrator prompt - scope, diff, constraints, expected output.
 2. Map changed symbols and the integration graph using Serena (`serena_get_symbols_overview` → `serena_find_symbol` with body=false → `serena_find_referencing_symbols`).
 3. Detect the project's architecture pattern from existing code, configs, AGENTS.md.
 4. Generate hypotheses about boundary violations, dependency inversions, hidden coupling.

.opencode/agents/flow-consistency-review.md

@@ -31,7 +31,7 @@ You are the consistency reviewer subagent for `rldyour-flow`. You are invoked on
 
 ## Review Focus
 
-- Naming: variables, functions, classes, modules, files, branches, environment variables — match project convention.
+- Naming: variables, functions, classes, modules, files, branches, environment variables - match project convention.
 - Style: indentation, formatting, comment density, JSDoc/docstring conventions, error message phrasing.
 - Imports: alphabetical / grouped / aliased per project rule; no cross-slice internal imports if FSD-like architecture; no circular imports.
 - Public API shape: matching nearby exports (named vs default, barrel files, index.ts pattern).
@@ -40,7 +40,7 @@ You are the consistency reviewer subagent for `rldyour-flow`. You are invoked on
 
 ## Workflow
 
-1. Read orchestrator prompt — scope, diff, constraints.
+1. Read orchestrator prompt - scope, diff, constraints.
 2. Establish baseline: read 3-5 nearby existing files in the same module/feature, plus AGENTS.md / Serena memories about conventions.
 3. Compare changed code against baseline.
 4. Report deviations as findings per `references/reviewer-protocol.md`.

.opencode/agents/flow-integration-review.md

@@ -40,7 +40,7 @@ You are the integration reviewer subagent for `rldyour-flow`. You are invoked on
 
 ## Workflow
 
-1. Read orchestrator prompt — scope, diff, constraints.
+1. Read orchestrator prompt - scope, diff, constraints.
 2. Use Serena (`serena_find_referencing_symbols`, `serena_search_for_pattern`) to trace cross-module references for changed contracts.
 3. For each contract change, check all touched layers.
 4. Report mismatch risks per `references/reviewer-protocol.md`.
@@ -54,5 +54,5 @@ Reply in Russian when user wrote in Russian.
 ## Anti-patterns
 
 - Modifying files.
-- Generic "check all integrations" findings — must point at concrete mismatch with code evidence.
+- Generic "check all integrations" findings - must point at concrete mismatch with code evidence.
 - Skipping migrations / backward-compatibility analysis when DB schema or public API changed.