Ghidra Headless Frida Evidence

Use this phase skill when planning must integrate externally captured Frida-derived dynamic evidence without making live Frida execution part of the supported ghidra-headless workflow.

The canonical contract is ./planning-brief.md. Use it to shape planning outputs and to audit generated artifacts for weakened Frida evidence requirements.

Phase Focus

This phase covers:

• imported Frida traces, logs, and hook context
• runtime-capture manifest consumption and evidence-manifest normalization
• provenance, replay, and verification expectations for dynamic observations
• explicit static-vs-dynamic conflict adjudication
• audit expectations for generated planning artifacts

Non-Negotiable Constraints

• Import-only scope. This skill does not execute Frida, provide live hook commands, or depend on GUI-driven capture activity.
• Runtime evidence must link back to a reviewable runtime-capture manifest from ../ghidra-headless-frida-runtime-injection/.
• Headless-only workflow. Imported evidence must remain compatible with headless Ghidra planning and review.
• Evidence-backed claims. Dynamic observations must trace to captured artifacts, provenance notes, or explicit verification surfaces.
• Observed claims stay distinct from inferred or unresolved claims.
• Conflicts with static analysis preserve both evidence surfaces until a reviewer records an explicit adjudication decision.
• Reproducible workflow expectations. Reviewers must be able to see what was captured, when it was captured, and how it maps back to the target.
• Reviewable Markdown outputs. Generated planning artifacts and findings remain inspectable as Markdown.
• Runtime artifacts stay under .work/ghidra-artifacts/ and are referenced explicitly rather than copied into the skill package.
• No downstream speckit extension or constitution change is required.

Required Inputs

• existing intake summary or normalized target context
• linked runtime-capture manifest and produced artifact references
• externally captured Frida evidence bundle
• provenance details for the capture, including source and timing context
• replay or verification expectations that a reviewer must confirm
• static evidence context when the runtime evidence might disagree with existing analysis
• known evidence gaps, ambiguities, or follow-up questions
• optional local overlays that only tighten the contract

How To Use This Skill

1. Fill in ./planning-brief.md with the imported Frida evidence bundle, linked runtime-capture manifest, provenance, and replay expectations.
2. Pass that brief into speckit as a file or inline paste.
3. Review the generated planning artifacts against the same Frida evidence checklist before treating them as ready for implementation.
4. Summarize the imported runtime outputs in ./templates/frida-evidence-manifest.md, keeping observed claims, inferred claims, and open questions separate.
5. If Frida-derived evidence conflicts with static analysis, record the conflict and an explicit reviewer decision instead of silently choosing one side.
6. If the plan introduces reusable normalization, translation, or manifest-generation scripts, route the follow-up through script review rather than weakening this import-only boundary.

Examples

• Frida evidence handoff example: ./examples/frida-trace-handoff.md
• Frida request rejection example: ./examples/contract-violation-example.md
• Frida conflict-preservation example: ./examples/frida-trace-contract-violation.md

Next Step Routing

• Use this phase after intake is stable and the planning request already has externally captured Frida-derived evidence plus a runtime-capture manifest.
• Return to the runtime-injection phase when the request still needs capture planning, reusable script selection, or runtime audit gates before evidence can be trusted.
• Return to generic evidence when the source is no longer Frida-specific or the dynamic capture format does not matter to the planning contract.
• Move to script authoring and review when the plan introduces reusable normalization, translation, manifest-generation scripts, or common-library behavior changes.
• Return to intake if target identity, provenance, or scope is still unstable.