shipshitdev Why
Investigate the motivation behind code. Companion to how. how answers
what the code does. why answers what forces led to its shape.
Historical context spreads across seven evidence categories. Enumerate available connectors at run time, query every available category in parallel, then synthesize with explicit confidence calibration. A null result is first-class evidence.
Contract
Inputs:
- A why-question about a chunk of code, pattern, feature, or named decision
Outputs:
- A confidence-separated, cited narrative with a coverage map
Creates/Modifies:
- None
External Side Effects:
- Read-only git,
gh, and available issue/docs/chat/observability connectors. No writes.
Confirmation Required:
- None
Delegates To:
- None. May spawn one investigator per evidence category, then one synthesizer.
Operating posture
- Evidence before narrative.
- Precision over polish. Quote and cite.
- Name the gaps.
- Hedge when evidence is indirect.
- Do not infer intent from code shape alone.
Read references/epistemics.md before synthesizing.
Step 1. Understand the target
Parse the question. If the target is vague, state the best guess from conversation context and proceed.
Step 2. Establish the code anchor
Before spawning investigators, gather file paths, line ranges, key symbols, recent commits, and PR numbers.
git blame -L <start>,<end> <file>
git log --follow -p -- <file>
git log --oneline -20 -- <file>
gh pr view <number> --json title,body,author,createdAt,mergedAt,labels,closingIssuesReferences,comments,reviews
Pass this seed to every investigator.
Step 3. Spawn parallel investigators
Discover available connectors (MCP servers, CLIs, APIs). Map each to one
category from references/evidence-categories.md.
Source control is always available through git and gh.
Launch one investigator per available category in a single message, on the fast cheap tier. Do not use a read-only sandbox that strips connectors. Investigators still write nothing.
Skip a category only with a written justification: no connector available, or the source is provably irrelevant. "Probably irrelevant" is not enough.
Each investigator gets references/investigator-prompt.md, its category playbook, the code anchor, and the user's question.
Step 4. Synthesize
Spawn one synthesizer on the strongest judgment tier. It gets all findings, including nulls, plus references/synthesizer-prompt.md and the epistemics file. Do not rewrite its confidence language.
Output format
The Question. Restate the ask.
The Code in Question. Paths, line ranges, symbols.
What We Found (direct evidence). Cited claims only.
What We Can Reasonably Infer. Hedged, with the inference chain.
Competing Hypotheses. When the record supports more than one story.
What We Don't Know. Specific gaps.
Sources Consulted. One line per category, including empties and skips.
If the why is a precursor to a change, convert findings into Preserve / Change / Avoid / Risk constraints.
Why procedure
Read why procedure when running this workflow. Apply the authorized scope and mode of this entry point to every step. Resolve other skills through this distribution’s active catalog; resolve resources relative to the installed skill directory.