Adds resume skill for continuing interrupted agent work

Author: sergeibbb

.claude/skills/resume-augint/SKILL.md

@@ -0,0 +1,84 @@
+---
+name: resume-augint
+description: Resume work from a previous Intent by Augment agent conversation that was interrupted (e.g., token limit, crash). Reads the prior agent's context and continues seamlessly. Only works in Intent by Augment workspaces.
+---
+
+# /resume-augint - Continue Interrupted Work (Intent by Augment)
+
+Pick up where a previous agent left off — no redundant questions, no re-investigation.
+
+## Usage
+
+```
+/resume-augint <agent-id-or-path>
+```
+
+**Accepted formats:**
+
+- Full agent ID: `agent-4bd8012e-f5f5-4d5a-b202-b03b31e61d06`
+- Workspace path: `.workspace/agents/agent-4bd8012e-....json`
+- Short ID prefix: `4bd8012e` (matches against known agents)
+- Agent name: `Coordinator 2` (fuzzy match against agent list)
+
+## Instructions
+
+### 1. Resolve the Agent
+
+- Use `list_agents(includeCompleted=true)` to find the agent
+- Match by ID, ID prefix, or name (case-insensitive)
+- If ambiguous, list matches and ask user to pick — this is the ONLY question allowed
+
+### 2. Gather Full Context (do ALL of these in parallel)
+
+- `get_agent_summary(agentId)` — quick overview of what the agent did
+- `read_agent_conversation(agentId)` — full conversation history with tool calls
+- If the agent has a task note: `read_note(taskNoteId)` — task details and acceptance criteria
+- `read_note("spec")` — current workspace spec and task checklist
+- `list_note_tasks("spec")` — current task completion status
+- `git status` and `git branch` — current repo state
+
+### 3. Build a Situation Report (internal, do NOT print verbose details)
+
+Synthesize from the gathered context:
+
+- **Original goal**: What was the agent trying to accomplish?
+- **Completed work**: What tasks/steps were finished?
+- **In-progress work**: What was the agent doing when it stopped?
+- **Failed/blocked items**: What errored or couldn't proceed?
+- **Current state**: Branch, uncommitted changes, task note statuses
+- **Remaining work**: What still needs to be done?
+
+### 4. Present a Brief Summary and Continue
+
+Print a SHORT summary (5-10 lines max):
+
+```markdown
+## Resuming: [Agent Name]
+
+**Goal:** [one-line description]
+**Done:** [completed items]
+**Stopped at:** [what was in progress when interrupted]
+**Remaining:** [what still needs to be done]
+**Current branch:** [branch] | **Uncommitted changes:** [yes/no]
+
+Continuing with [next action]...
+```
+
+Then **immediately start working** on the next incomplete item. Do NOT ask for permission to continue — the user invoked `/resume` precisely because they want you to pick up and go.
+
+### 5. Execution Rules
+
+- **Do NOT re-do completed work** — trust the previous agent's results unless evidence of failure
+- **Do NOT re-ask questions** the previous agent already answered or investigated
+- **Do NOT change approach** unless the previous approach clearly failed — continue the same strategy
+- **DO verify** the current state matches expectations (e.g., branches exist, files are as expected)
+- **DO check** if delegated sub-agents completed or failed, and handle accordingly
+- **DO update** task statuses in the spec as you complete items
+
+## Anti-Patterns
+
+- Asking "What would you like me to do?" — the previous conversation already established this
+- Re-running investigation steps that the previous agent already completed
+- Printing the entire previous conversation back to the user
+- Changing the plan without evidence that the original plan was wrong
+- Ignoring failed delegated agents — check their status and retry or handle manually