agent-mockingbird 0.0.1

package/runtime-assets/workspace/.agents/skills/config-auditor/SKILL.md

@@ -0,0 +1,25 @@
+---
+name: config-auditor
+description: Audit Agent Mockingbird config for drift, risky values, and operability issues, then suggest minimal safe patches.
+---
+
+# config-auditor
+
+Use this skill to evaluate config quality before changing it.
+
+## Workflow
+
+1. Fetch current config with `config_manager` using `get_config`.
+2. Inspect runtime settings for:
+   - timeouts that are too low for current providers
+   - memory settings that can cause noisy retrieval
+   - cron defaults that may cause retry storms
+   - invalid model/provider references
+3. Propose explicit patch candidates.
+4. Apply only when the user asks to apply.
+
+## Rules
+
+- Favor small, isolated changes.
+- Explain expected impact of each proposed patch.
+- Never apply multiple unrelated config changes in one patch unless requested.

package/runtime-assets/workspace/.agents/skills/config-editor/SKILL.md

@@ -0,0 +1,24 @@
+---
+name: config-editor
+description: Safely read and update Agent Mockingbird runtime configuration with hash-aware, smoke-tested writes.
+---
+
+# config-editor
+
+Use this skill for runtime configuration updates.
+
+## Required workflow
+
+1. Call `config_manager` with `action: "get_config"` and capture `hash`.
+2. Build a minimal `patch`.
+3. Call `config_manager` with:
+   - `action: "patch_config"`
+   - `expectedHash` from step 1
+   - `runSmokeTest: true`
+4. If a hash conflict occurs, refresh with `get_config` and retry once with a fresh hash.
+
+## Rules
+
+- Prefer `patch_config` over `replace_config`.
+- Keep patches narrow and reversible.
+- Do not attempt to modify smoke-test policy fields unless explicitly instructed by the user.

package/runtime-assets/workspace/.agents/skills/cron-manager/SKILL.md

@@ -0,0 +1,57 @@
+---
+name: cron-manager
+description: Create, inspect, update, run, and delete Agent Mockingbird cron jobs using cron_manager with the 3-mode model (background, conditional_agent, agent).
+---
+
+# cron-manager
+
+Use this skill when users want to manage scheduled jobs in Agent Mockingbird.
+
+## Cron mode model
+
+- `background`: execute workspace module at `conditionModulePath`; no agent step.
+- `conditional_agent`: execute workspace module at `conditionModulePath`; module can request agent invocation with `invokeAgent.shouldInvoke === true`.
+- `agent`: agent runs every execution.
+
+## Required fields by mode
+
+- `background`
+- requires: `conditionModulePath`
+- should include: `conditionDescription` when a short Job Details summary is useful
+- should not include: `agentPromptTemplate`
+
+- `conditional_agent`
+- requires: `conditionModulePath`
+- should include: `conditionDescription` (1-2 sentence plain summary for Job Details UI)
+- `agentPromptTemplate` is optional fallback when module does not provide `invokeAgent.prompt`
+
+- `agent`
+- requires: `agentPromptTemplate`
+- should not include: `conditionModulePath`
+
+## Safe workflow
+
+1. Call `cron_manager` with `action: "describe_contract"` and `action: "list_jobs"` first.
+2. Prefer `action: "upsert_job"` with stable `job.id` for idempotent create/update.
+3. For create/update, send minimal fields and preserve unrelated job settings.
+4. After creating/updating, verify with `action: "get_job"`.
+5. Optionally smoke test with `action: "run_job_now"` and inspect `list_instances` / `list_steps`.
+6. For deletions, confirm intent before `action: "delete_job"`.
+7. Prefer pausing via `action: "disable_job"` when the user wants to stop runs without deleting history.
+
+## Patterns
+
+- Create recurring background job:
+- `scheduleKind: "every"`, `everyMs`, `runMode: "background"`, `conditionModulePath`, optional `conditionDescription`.
+
+- Create cron-expression conditional job:
+- `scheduleKind: "cron"`, `scheduleExpr`, `runMode: "conditional_agent"`, `conditionModulePath`, `conditionDescription`, optional `agentPromptTemplate`.
+
+- Create one-shot agent job:
+- `scheduleKind: "at"`, `atIso`, `runMode: "agent"`, `agentPromptTemplate`.
+
+## Rules
+
+- Prefer `upsert_job` with stable IDs over delete+create.
+- Use `enable_job` / `disable_job` for quick pause/resume instead of delete+recreate.
+- Keep payloads small and explicit; avoid storing secrets in payload unless user explicitly requests it.

package/runtime-assets/workspace/.agents/skills/memory-ops/SKILL.md

@@ -0,0 +1,120 @@
+---
+name: memory-ops
+description: Use Agent Mockingbird memory effectively for persistent context across sessions.
+---
+
+# memory-ops
+
+Use Agent Mockingbird memory effectively for persistent context across sessions.
+
+## Memory Tools
+
+| Tool | Purpose |
+|------|---------|
+| `memory_search` | Find relevant prior context by semantic query |
+| `memory_get` | Read specific memory file by path |
+| `memory_remember` | Persist a durable fact, decision, or context |
+
+## Workflow
+
+### 1. Always Search First
+
+Before making decisions or asking clarifying questions, search memory:
+
+```
+memory_search(query="project architecture decisions", maxResults=5)
+memory_search(query="user preferences for tests", maxResults=3)
+```
+
+Use semantically specific queries first, then broaden carefully. Current runtime defaults are `maxResults=4` and `minScore=0.35`.
+
+#### Query steering (important)
+
+- Prefer concrete entities over vague categories.
+- Include relationship terms the memory likely contains.
+- Try 2-4 reformulations before concluding "not found".
+
+Examples:
+
+```
+# weak (often misses lexical filter)
+memory_search(query="family")
+
+# stronger
+memory_search(query="daughter name")
+memory_search(query="Lucy Lee")
+memory_search(query="wife Tiffany")
+memory_search(query="who is my daughter Lucy")
+
+# broader adjacent lookup (portfolio -> metals/silver)
+memory_search(query="portfolio allocation metals silver")
+memory_search(query="investments including precious metals")
+```
+
+If zero results:
+
+1. Retry with exact entities (`Lucy Lee`, `Tiffany`, `silver`).
+2. Retry with relation/domain words (`daughter`, `wife`, `spouse`, `child`, `portfolio`, `metals`).
+3. Lower threshold once (`minScore=0.2`) for recall probe.
+4. If still empty, confirm memory was actually written and indexed.
+
+### 2. Validate with Get
+
+Memory files live in `MEMORY.md` or `memory/*.md`. Use `memory_get` to read full context around a snippet:
+
+```
+memory_get(path="memory/decisions.md", from=1, lines=50)
+```
+
+### 3. Persist What Matters
+
+Use `memory_remember` for durable information that should survive session boundaries:
+
+**Good candidates:**
+- Project-level decisions and rationale
+- User preferences and constraints
+- Recurring patterns or conventions discovered
+- Critical context about third-party integrations
+
+**Parameters:**
+
+| Param | Purpose | Example |
+|-------|---------|---------|
+| `content` | The fact/decision to remember | "Use Bun, not Node.js" |
+| `topic` | Category for organization | "conventions", "decisions" |
+| `entities` | Related identifiers | ["bun", "runtime"] |
+| `confidence` | Certainty level 0-1 | 0.9 for verified facts |
+| `source` | Who provided this | "user", "assistant", "system" |
+| `supersedes` | IDs of outdated memories to replace | ["mem_abc123"] |
+
+### 4. Supersede, Don't Duplicate
+
+When information changes, include `supersedes` to link the replacement:
+
+```
+memory_remember(
+  content="Testing framework: bun test (switched from vitest)",
+  topic="conventions",
+  supersedes=["mem_old_testing_config"]
+)
+```
+
+## Memory Modes
+
+Runtime config `runtime.memory.toolMode` controls behavior:
+
+| Mode | Behavior |
+|------|----------|
+| `hybrid` | Prompt context + tools available |
+| `inject_only` | Prompt context only, tools disabled |
+| `tool_only` | Tools only, no prompt injection (default) |
+
+Adjust usage accordingly — in `inject_only`, rely on existing memories in prompt context.
+
+## Anti-Patterns
+
+- **Don't** use only abstract terms as your only query; follow with specific entities/adjacent terms
+- **Don't** memorize ephemeral state (current task progress, temp files)
+- **Don't** write memories for information already in code/docs
+- **Don't** create redundant memories — search first, then supersede
+- **Don't** use low confidence (<0.5) for critical decisions

package/runtime-assets/workspace/.agents/skills/runtime-diagnose/SKILL.md

@@ -0,0 +1,25 @@
+---
+name: runtime-diagnose
+description: Diagnose runtime failures (timeouts, send failures, model/provider issues) and recommend config fixes.
+---
+
+# runtime-diagnose
+
+Use this skill when the runtime is reachable but requests fail, stall, or time out.
+
+## Workflow
+
+1. Read current config and hash via `config_manager` (`get_config`).
+2. Correlate failures with settings:
+   - `runtime.opencode.timeoutMs`
+   - `runtime.opencode.promptTimeoutMs`
+   - `runtime.opencode.runWaitTimeoutMs`
+   - `runtime.runStream.*`
+3. Propose one minimal fix at a time.
+4. If asked to apply, use `patch_config` with `expectedHash` and `runSmokeTest: true`.
+
+## Rules
+
+- Prefer timeout tuning before model/provider swaps.
+- Avoid broad config replacement during incident response.
+- After each applied change, verify the smoke test result in the response.

package/runtime-assets/workspace/AGENTS.md

@@ -0,0 +1,56 @@
+# Agent Mockingbird Runtime Agent Guide
+
+You are operating as a runtime assistant inside a Agent Mockingbird workspace.
+
+## Priorities
+
+1. Follow explicit user instructions.
+2. Use configured tools and skills instead of guessing.
+3. Keep changes minimal and reversible.
+
+## Tool Usage
+
+For independent reads, searches, listings, and other non-dependent tool calls, prefer the `batch` tool so they run in parallel.
+
+Do not use `batch` for ordered mutations or steps that depend on earlier tool output.
+
+## Runtime Skills
+
+Use workspace skills from `.agents/skills` when relevant:
+
+- `config-editor` for safe config patching with hash + smoke test.
+- `config-auditor` for drift/risk audits and minimal patch proposals.
+- `runtime-diagnose` for runtime timeout/provider/model incident triage.
+- `memory-ops` for memory status/sync/reindex/retrieval validation.
+- `cron-manager` for cron job lifecycle operations.
+
+## Cron Behavior
+
+When using cron tools:
+
+1. Start with `cron_manager` `action: "describe_contract"` for current mode requirements.
+2. Prefer `action: "upsert_job"` with explicit `job.id` for idempotent job management.
+3. For `background` and `conditional_agent`, use `conditionModulePath` (workspace module file). For `conditional_agent`, include `conditionDescription` for the Job Details summary.
+4. Use `agentPromptTemplate` as optional fallback prompt; per-run overrides can come from `invokeAgent.prompt`.
+5. Use `run_job_now` + `list_instances` + `list_steps` to validate behavior after create/update.
+6. Use `disable_job` to pause without deleting, and `enable_job` to resume.
+
+## Memory Behavior
+
+Durable memory lives in workspace markdown (`MEMORY.md` and `memory/*.md`).
+
+When memory tools are available:
+
+1. Use `memory_search` first.
+2. Validate details with `memory_get` before relying on them.
+3. Persist durable facts/decisions with `memory_remember`.
+4. For people/relationship recall, start with concrete terms (for example: daughter, spouse, partner, child, parent, names) instead of only broad words.
+5. When search misses, reformulate once using concrete entities/relationship words before concluding no memory exists.
+
+Respect runtime memory mode:
+
+- `hybrid`: prompt memory + tools
+- `inject_only`: prompt memory only
+- `tool_only`: tools only
+
+When replacing older memory content, include `supersedes` where possible.

package/runtime-assets/workspace/MEMORY.md

@@ -0,0 +1,4 @@
+# Memory Index
+
+This file is part of the runtime workspace bundle.
+Store durable notes in `memory/*.md` and keep this file as a high-level index.

package/scripts/build-release-bundle.sh

@@ -0,0 +1,66 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+DIST_DIR="${ROOT_DIR}/dist"
+STAGE_DIR="${DIST_DIR}/release-stage"
+
+VERSION="${1:-}"
+if [[ -z "${VERSION}" ]]; then
+  echo "usage: $0 <version>"
+  echo "example: $0 v0.1.0"
+  exit 1
+fi
+
+if ! command -v git >/dev/null 2>&1; then
+  echo "git is required"
+  exit 1
+fi
+
+if ! command -v bun >/dev/null 2>&1; then
+  echo "bun is required"
+  exit 1
+fi
+
+mkdir -p "${DIST_DIR}"
+rm -rf "${STAGE_DIR}"
+
+ARCHIVE_PATH="${DIST_DIR}/agent-mockingbird-${VERSION}.tar.gz"
+CHECKSUM_PATH="${ARCHIVE_PATH}.sha256"
+PREFIX="agent-mockingbird-${VERSION}"
+
+echo "Building vendored OpenCode app + standalone runtime..."
+(
+  cd "${ROOT_DIR}"
+  bun run build
+  bun run build:bin
+)
+
+mkdir -p "${STAGE_DIR}/${PREFIX}"
+
+echo "Exporting tracked files..."
+git -C "${ROOT_DIR}" archive --format=tar HEAD | tar -x -C "${STAGE_DIR}/${PREFIX}"
+
+echo "Refreshing built dist artifacts..."
+rm -rf "${STAGE_DIR:?}/${PREFIX}/dist"
+cp -R "${ROOT_DIR}/dist" "${STAGE_DIR}/${PREFIX}/dist"
+rm -rf "${STAGE_DIR:?}/${PREFIX}/dist/release-stage"
+
+test -f "${STAGE_DIR}/${PREFIX}/dist/agent-mockingbird"
+test -f "${STAGE_DIR}/${PREFIX}/dist/drizzle/meta/_journal.json"
+test -f "${STAGE_DIR}/${PREFIX}/dist/app/index.html"
+test -f "${STAGE_DIR}/${PREFIX}/drizzle/meta/_journal.json"
+
+echo "Packing release bundle..."
+tar -C "${STAGE_DIR}" -czf "${ARCHIVE_PATH}" "${PREFIX}"
+
+(
+  cd "${DIST_DIR}"
+  sha256sum "agent-mockingbird-${VERSION}.tar.gz" > "agent-mockingbird-${VERSION}.tar.gz.sha256"
+)
+
+rm -rf "${STAGE_DIR}"
+
+echo "Created:"
+echo "  ${ARCHIVE_PATH}"
+echo "  ${CHECKSUM_PATH}"