impact

tsp:impact

Reverse-map code artifacts (files, diffs, symbols) to TSP nodes via tsp.impact.map. Useful before refactors, during review, and before merging work that might cross semantic boundaries.

V1 ranking sources, descending by inherent confidence:

• recorded_code_ref (~0.9): a node's record_result previously reported the touched path.
• keyword_match (~0.45-0.6): the artifact's path / symbols / excerpt overlap with the node's title, intent, or scope.
• edge_neighbor (~0.35): one hop along depends_on from a candidate.

The mapping improves over time because implement and handoff record touched files; cold-start results are lower-confidence and lean on keyword + edge traversal.

Capability

impact <path> — given a file path (or diff, or set of symbols), list the TSP nodes most likely to govern that code, ranked by confidence, with the in-scope items that apply and the risks of touching them.

Inputs

• path: a file path (required if no diff or symbols are provided).
• diff: a unified diff (optional; bounded at 40k chars).
• symbols: explicit symbols / function names (optional).
• mode: pre_refactor / pre_review / general (default).
• plan_id and branch_name: optional; default to session context.

Required MCP tools

• tsp.context.get — resolve session context.
• tsp.impact.map — the reverse mapping.
• tsp.node.execution_context (optional) — fetch full intent and acceptance criteria for the top candidate so the harness can render them inline.

Local filesystem responsibilities

The harness:

1. Reads the file at path (or the diff for the current branch) and captures a bounded excerpt.
2. Optionally extracts symbols (function / class names) using its own parser or simple regex.
3. Sends only bounded content to MCP: 20 artifacts max, 20k content chars per artifact, 40k diff chars.

Tool sequence

1. tsp.context.get
2. (Optional) Local: read the file or capture the diff; extract symbols.
3. tsp.impact.map(
     selector=<plan>,
     artifacts=[CodeArtifactInput(path=..., content_excerpt=..., symbols=...)],
     mode=<pre_refactor|pre_review|general>,
     max_candidates=10,
     include_dependency_neighbors=true,
   )
4. (Optional, top candidate): tsp.node.execution_context(selector=<top_node>)
5. Render the candidate list with rationale, governing scope, and risks.

Output format

🎯 Impact map for <path or "current diff">
   plan: <plan_id>@<branch>

Candidates (top <N>):
  1. <node_id> — <title>  [<relation>, confidence <0.0-1.0>]
       rationale: <one-line rationale>
       in-scope: <comma-separated scope items or "—">
       risks: <comma-separated risks or "—">
  ...

Dependency neighbors:
  - <edge_id>: <from> -> <to> (<type>)

Suggested checks:
  - /tsp:align <node-id>
  - /tsp:impact current diff

For empty results, the skill surfaces the warning verbatim and recommends running /tsp:nodes.search with terms from the artifact.

Blocking conditions

Condition	Skill behavior
No code refs recorded anywhere	Warn that confidence is low; the keyword + edge fallback still produces candidates but the user should verify.
Artifact exceeds 20k content cap	Trim to the function bodies and import lines; fall back to symbols-only if still over.
NO_CANDIDATES	Show the warning verbatim and ask the user to identify the governing node manually, then offer to record it.
Multiple equally likely candidates	Render all of them in rank order; pick the top one for the optional execution-context enrichment.

State writes

This skill never writes plan state. It surfaces information; the user or /tsp:implement writes status separately.