--- name: pm-investigation-585b8a0968a7 description: | Create or manage an investigation workflow packet. Use when the user needs to explore an unknown, test hypotheses, or answer a question whose output is knowledge rather than code. Structures the work around hypotheses, evidence collection, and decision-ready conclusions. --- # pm-investigation ## Purpose Create or manage an investigation workflow packet. Investigations explore unknowns, test hypotheses, and produce decision-ready conclusions. The output is knowledge, not code. This skill emphasizes structured hypotheses, evidence collection, and clear recommendations. ## Required Reads 1. `.agents/workflows/INDEX.md` -- understand lifecycle states and packet structure. 2. `.agents/workflows/dossier-template.md` -- the dossier format to follow. 3. `.agents/pm/pm-current-state.md` -- check for related active work and known risks. 4. Relevant PM units for the area under investigation (via pm-load). ## Procedure ### Creating a new investigation packet 1. **Choose the workflow ID.** Use the prefix `iv_` followed by a short descriptive slug. Example: `iv_memory_leak_api_server`. 2. **Create the packet directory.** Path: `.agents/workflows/draft/iv_/`. 3. **Write `dossier.md`.** Copy the structure from `dossier-template.md` and fill in: - Metadata: set Workflow Type to "investigation", Lifecycle State to "draft". - Objective: state the question being answered. Frame it as a question, not a task. Example: "Why does the API server's memory grow by 50MB/hour under steady load?" - Background/Context: summarize what is already known, including symptoms, prior attempts, and relevant data points. - Contract: define what a sufficient answer looks like. Example: "Root cause identified with evidence, and at least two remediation options evaluated." - Acceptance Criteria: list the specific deliverables. Typical set: - Hypotheses listed with supporting and contradicting evidence. - Root cause identified or top candidates ranked by likelihood. - Recommended action with rationale. - Key evidence preserved in artifacts. - Failure Policy: define when to stop investigating. Default: "If no progress after 3 evidence-gathering cycles, document findings so far and recommend external escalation." - Scope: list the files, logs, systems, or data sources to examine. - Plan of Record: structure the investigation as: (a) form hypotheses, (b) design evidence-gathering steps, (c) collect evidence, (d) evaluate and narrow, (e) conclude. - Verification Strategy: describe how conclusions will be validated (e.g., reproduce the issue, confirm fix in staging). - Open Questions: seed with the initial unknowns. 4. **Write `run-log.md`.** Initialize from `run-log-template.md`. Investigation run-logs should record each hypothesis tested, evidence found, and whether the hypothesis was supported or refuted. 5. **Write `handoff.md`.** Initialize from `handoff-template.md`. Set "Next Exact Step" to the first evidence-gathering action. 6. **Write `artifacts.md`.** Initialize from `artifacts-template.md`. Investigations should index: data captures, log excerpts, benchmark results, and any reproduction scripts. 7. **Add a row to `.agents/state/active-workflows.md`.** That file is the workflow tracking table, and it is preserved across upgrades. Do not open a second table in `.agents/workflows/INDEX.md`: that file is framework reference content, an upgrade refreshes it, and anything recorded there is lost. ### Managing an existing investigation packet 1. **After each evidence-gathering step**, append a run-log entry: hypothesis under test, evidence collected, verdict (supported / refuted / inconclusive), and remaining hypotheses. 2. **Update `handoff.md`** with current hypothesis ranking and next evidence step. 3. **When concluding**, write a summary in `artifacts.md` with: final answer, confidence level, key evidence references, and recommended next actions. 4. **Move the packet to done** only when the acceptance criteria are met. If the question cannot be fully answered, move to done with partial findings documented, or to failed if no useful progress was made. ## Output - `.agents/workflows/draft/iv_/dossier.md` - `.agents/workflows/draft/iv_/run-log.md` - `.agents/workflows/draft/iv_/handoff.md` - `.agents/workflows/draft/iv_/artifacts.md` ## Completion Criteria - All four packet files exist and follow their respective templates. - The objective is framed as a question. - At least two hypotheses are listed in the dossier or run-log. - The contract defines what a sufficient answer looks like. - The failure policy includes a stop condition to prevent unbounded investigation. ## Error Handling - If the investigation scope is too broad (more than 5 hypotheses spanning unrelated systems), recommend splitting into multiple investigation packets. - If evidence is inaccessible (e.g., production logs unavailable), document the gap in the dossier under Constraints and adjust the plan to work with available data. - If the investigation reaches the failure policy threshold, write up partial findings, move the packet to failed, and persist any useful negative knowledge to PM via pm-update. --- Source attribution: AMDResearch Original project: https://github.com/AMDResearch/omniprobe/blob/913ae319f727f8014e4db442a80e84915c8cf3a0/.agents/skills/pm-investigation/SKILL.md License: MIT License information: https://opensource.org/license/mit Original upstream instructions; downloaded from Mega AI Hub. ## Upstream license notice ------------------------------------------------------------------------------- MIT License Copyright (c) 2026 Advanced Micro Devices, Inc. All Rights Reserved. Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. ------------------------------------------------------------------------------- # Mega AI Hub resource metadata Creator / publisher: AMDResearch Original project: https://github.com/AMDResearch/omniprobe/blob/913ae319f727f8014e4db442a80e84915c8cf3a0/.agents/skills/pm-investigation/SKILL.md Attribution basis: Source repository record Fields of work: Data & Analytics, Operations, Project & Program Management Roles: Business Process Analyst, Data Analyst, Project Manager Specialties: Specialty not specified Tools: Tool not specified Country applicability: Country not specified Classification basis: Source catalog labels Instruction SHA-256: af6dc616e07cfea1289cbfecf54dcf18f9e1923407f06d57a89889754ad35d6b